You are on page 1of 26

2019-2020 staff MSc project list

See staff pages at https://www.essex.ac.uk/departments/computer-science-and-electronic-engineering/people/academic to help guide you in your project selection

Computer Science
Masters courses Telecommunications and data communications Masters courses
MSc Advanced Computer MSc Computer Networks and Security
Science MSc Electronic Engineering
MSc Artificial Intelligence MSc Advanced Communication Systems
MSc Big Data and Text
Analytics
MSc Computer
Engineering
MSc Internet of Things
MSc Intelligent Systems
and Robotics
MSc Computer Games

Staff member Project title Project outline Status MSc Programme/


Prerequisites

Abolghasemi, Vahid Compressive object Object tracking has been always a challenging task in computer vision. Recent researches have shown that interesting theory of Available Any degree
tracking (video) compressive sensing (CS), can significantly improve the performance of object tracking (pedestrian, vehicle, etc.) in video and also reduce
the processing time. Thanks to CS theory, one can reduce the number of pixels in video signals while maintaining quality needed for
recovering the trace of an objects. A more challenging case would be multi-object tracking using compressed sensing and sparse signal
representation. There is open research in using CS for tracking as well as suitable feature representation in this application.

Smartphone-based Wide range of smartphone use in our daily life is obvious to everyone. The aim of this research could be to measure the breathing activity Available Any degree
Breathing Activity of human using a wearable device and a smartphone. Breathing-related biomedical signals are breathing sound or chest movement
Monitoring by using a which can be obtained by wearable
Wearable Device devices. It could be extended/combined to/with other biomedical signals such as ECG as well. An algorithm is proposed and written as a
mobile app for human biomedical signal monitoring. It then processes the measured data from the sensor and applies different steps e.g.
preprocessing, feature extraction, classification, etc. The output of the algorithm is a measure of degree of healthiness and could be sent
to a doctor if needed.

Compressive Sensing in Energy efficient data gathering is very important in wireless sensor networks. Considering the fact that wireless sensor nodes are often Available Any degree
Wireless Sensor Networks deployed densely (meaning that they are sparse), and knowing that compressed sensing is a powerful tool for sparse data reconstruction,
an interesting connection between the two subjects can be established. Some recent researches have shown the advantage of
compressive sensing for wireless sensor networks. However, it is still an open research field and could lead to interesting results when CS
meets WSN.

Image classification using This is a general topic for image classification. Different case studies such as natural images, face, OCR, biomedical images, MRI, artifact Available Any degree
dictionary learning removal, etc. could be selected. The main suggestion is to use dictionary learning techniques that has shown to be powerful for
discrimination and classification with the very high performance. Another advantage is that, dictionary learning can be simultaneously
used for denoising and classification. One of the most famous dictionary learning algorithms is K-SVD. It was initially used for face
recognition. The method sometimes called Sparse Representation Classification (SRC). I can provide related references for you to study
more on this and find out interesting applications using this technique.

Fetal ECG extraction from Real-time extraction of ECG signals from compressed (and sometimes mixed/noisy) data is a great challenge. It is recently shown that Available Any degree
multichannel abdominal combining compressed sensing (CS) and intendent component analysis (ICA) can readily help to achieve this goal. So, there are different
ECG using compressive possibilities of combining blind source separation (BSS) methods with compressed sensing in order to extract useful information from the
sensing abdominal ECG signals. The ultimate aim is reaching less processing time and higher reconstruction quality.

Welcome a student Available Any degree


suggestion on a project in
the area of signal and
image processing,
particualrly based on
compressed sensing and
dictionary learning

Al-Naday, Mays Malware analysis using The project will investigate the use of machine learning approaches to identify malware behaviour in different edge devices. The project Available Any degree
Machine Learning (with requires having programming skills, particularly in python.
BT)

Network Function NFV and Software Defined Networking (SDN) are critical paradigms to facilitate 5G and beyond networks. Existing research work has Available Any degree
Virtualisation (with BT) focus on NFV and SDN independently; however, their integration and co-operation is a key factor in realising their full potential. The
project will investigate solutions for efficient communication of Virtual Network Functions when connected over a Software Defined
Network.

Amorim, Renato Student’s own suggestion It is open to students to suggest topics for projects related to data clustering. Please make an appointment to discuss possibilities. Available Any degree
for projects related to
data clustering

Malware taxonomy Currently, over a million malwares are being released on a daily basis. With such a high amount of malware being released, it is desirable Available Any degree
to create automatic methods to determine how many families of malware there are, as well as assign each malware to a particular
family. Such method would allow anti-virus companies to write behavioural signatures to detect future malware of a particular family
with low false positives and false negatives. In this project a student will have to gather malware behavioural data using a honeypot and
analyse this data.

Exploring consensus A large number of clustering algorithms have been introduced in the literature. A single algorithm may produce a set of different Available Any degree
clustering clusterings if run with different parameters (and in some cases, even if run with the same parameters). In this project the student will
explore methods that generate a single clustering given a set of clusterings. This final single clustering may or may not be part of the
initial set, making this matter way beyond the usual capabilities of clustering validation indices.

Finding homogeneity in Nowadays the internet offers a large number of texts which are relatively easy to obtain, from ordinary webpages to scientific papers. Available Any degree
texts Given the high amount of such texts being released on a daily basis it is simply impossible to manually categorise them. In this project the
student will investigate methods to automatically generate clusters containing related (homogeneous) texts. This project is co-supervised
by Aline Villavicencio, and the student may have access to (anonymised) data from the Essex County Council.

Dealing with partially Classification algorithms usually require a high proportion of labelled samples. However, manually labelling modern (and by consequence Available Any degree
labelled data sets large) data sets tend to be a very expensive and time consuming task. In this project students will explore methods that can be used to
increase the proportion of labelled objects in a data set.
Finding the number of Thanks to modern computer systems we can easily access vast amounts of data. Unfortunately, most of this data is unlabelled, and often Available Any degree
clusters in data sets we don’t even know how many different labels there should be. In this project the student will explore various methods that can be used
to find the number of clusters in data sets, as well as apply these to a particular problem (to be discussed).

Andreu-Perez, Javier Drone (UAV) Imaging for Thermographic imagery captured by drones has become a vital tool for the inspection and maintenance of solar farms: A wide range of Available MSc Advanced Computer
Renewable Energy Assets photovoltaic module issues are indicated by heating and are clearly identifiable in the thermal spectrum. However, geolocating modules Science, MSc Computer
in thermal imagery can be challenging for commonly used structure from motion techniques, due to their low resolution, and often Engineering, MSc
extremely homogeneous representation. This project’s aim is to investigate and implement and evaluate the use of planar homography Artificial Intelligence,
to perform camera pose estimation in order to extract an accurate model of a solar farm from thermal imagery. For this you will have to MSc Intelligent Systems
make use of computer vision techniques and the OpenCV package. If you have any question about this project contact and Robotics, MSc Big
javier.andreu@essex.ac.uk. Data and Text Analytics,
MSc in Data Science.

Build and design your In this project the student will have to be able to develop a bio-sensor that will collect spectroscopy data (visible and non-visible). This Available MSc Electronic
smart optical bio-sensor could be used for the monitoring of chemical conditions of tissues, that could be from any bio-material. The student should be able to Engineering, MSc
(Electronics and 3D have some knowledge and interest of electronics. The student will be able to design their own case for the sensor using the 3D printing Intelligent Systems and
printing) facilities of the school. If you have any question about this project contact javier.andreu@essex.ac.uk. Robotics

Bio-engineering project This project focus on determining the neural mechanisms of an underlying cognitive processes and develop models to either analyse and Available MSc Advanced Computer
for Cognitive visualise it. The student will be provided with data from a UEssex study. This student could use biomedical processing techniques and Science, MSc Computer
Neuroscience Machine Learning in their project. This project is particularly interesting for students with interest in bio-engineering or with interests in Engineering, MSc
Applications. the area . If you have any question about this project contact javier.andreu@essex.ac.uk. Artificial Intelligence,
MSc Intelligent Systems
and Robotics, MSc Big
Data and Text Analytics,
MSc in Data Science.

eXplanaible Artificial Artificial Intelligence, Machine Learning and Deep Learning have become an increasingly popular tool for solving complex tasks. In this Available MSc Advanced Computer
Intelligence (XAI) project, the student aims at experimenting and developing new ways of using deep learning in scenarios where expressiveness and Science, MSc Computer
reproducibility are more important than performance per se. This entail adding some novel processing framework or logic into the Engineering, MSc
artificial intelligence process which will help end-users to deciphering the black-box. During this project, the student will acquire valuable Computer Networks and
experience in the existing machine/deep learning platform and an in depth theoretical understanding about their components and Security, MSc Artificial
architecture. For this project, the student should be familiar with the foundations of machine learning and AI, or has previously coursed a Intelligence, MSc
module dealing with this subject. If you have any question about this project contact javier.andreu@essex.ac.uk. Intelligent Systems and
Robotics, MSc Big Data
and Text Analytics, MSc
in Data Science, MSc
Computer Games.

Decoding and Making a successful video game is hard and it is even harder to make players retain their interest. Studying User Experience (UX) has Available MSc Advanced Computer
Understanding the become a very important science in the development of video games. UX is about understanding the gamer’s cognition: understanding Science, MSc. Advanced
Gamer's cognition. human capabilities and limitations to anticipate how a game will be perceived and their responses. For this project we have already Web Engineering, MSc
collected a large battery of data from gamers including professional ones that dedicate their lives to partificate in Esports. In this project Computer Engineering,
the student will have to analyse gamer's data , get some insight into the different stimulus from the game, gaming situation and the MSc Artificial
effect that they have in the game. It is desirable that the student has some knowledge or interests in computer vision and deep learning. Intelligence, MSc Big
If you have any question about this project contact javier.andreu@essex.ac.uk. Data and Text Analytics,
MSc in Data Science, MSc
Intelligent Systems and
Robotics, MSc Cloud
Computing, MSc
Computer Networks and
Security.

Student’s own In this project a student will have the opportunity to experience the world of quantum computing with one of the existing simulators that Available MSc Advanced Computer
suggestions project on are in the cloud, for instance IBM Q Experience. In this project the student can generate a quamtum circuit model for any specific Science, MSc Computer
Quantum Computing customized computation by applying quamtum gates. There are different GUI iditors as well as python apis for doing this. The student Engineering, MSc
can use this project to learn about an exciting new technology that will revolutionise computer science in the next 5 to 10 years. The Artificial Intelligence,
student can also make use of any of the existing quamtum algorithms usch as Grover's, Shor's or other algorithims. Hybrid MSc Intelligent Systems
quantum/classical algorithms can be also considered. It would be desirable that for this project the student has already read some of the and Robotics, MSc
literature concering Quantum Computing, although this is not required. Profiency in python would be desirable to be able to interact with Electronic Engineering,
the Quantun simulator as well as a capacity in understanding algorithms. If you have any question about this project contact MSc Data Science, MSc
javier.andreu@essex.ac.uk. Computer Networks and
Security, MSc Electronic
Engineering.

Develop an Emotional AI To enable a closest interaction and effective collaboration between machines and humans, machines must be able to interpret the Available MSc Advanced Computer
agent using Affective emotions and states of the human beings in the enviroments. In this project the student can build an application that will automatical Science, MSc Computer
Computing. estimate the emotional conditions of humans based on affective computing techniques derived from computer vision, face analysis bio- Engineering, MSc
sensing or eye tracking. Students can make use of some of the existing databases in the internet, and can also develop their own Artificial Intelligence,
scenarions and case studies. The project does require the use of machine learning / deep learning and signal processing. If you have any MSc Intelligent Systems
question about this project contact javier.andreu@essex.ac.uk. and Robotics, MSc Big
Data and Text Analytics,
MSc in Data Science.

AI Explainability – AI Explainability – Argumentation Layer for resource scheduling: Available Knowledge/Skills:


Argumentation Layer for Our scheduling solution is responsible for allocating work to field technicians using a constraint-based model. Programming skills in
resource scheduling To do so it is optimising a multi-objective function using Local Search algorithms maximising allocation of tasks and minimising costs Java, Restful Web
(industrial project with (travel, quality of schedule, etc.). Services
BT, co-supervised by Dr. However to some extends this solution works as a black box that:
Raphael Dorne) - do not provide any explanation back to the end-user to justify any choice made,
- do not offer any mean to interact (argument) with the scheduler even though a lot of information is readily available inside or can be
easily derived from.

Hence the purpose of this project is to build an argumentation layer around this existing field force scheduling solution as depicted below.
This argumentation layer will be responsible for interacting with end-users in a human manner where proper conversations are expected.
Chatbots and Restful APIs technologies will be used to build this component.

Knowledge/Skills:
Programming skills in Java, Restful Web Services
Angelopoulos, Nicos Analysis and predictive This project will consist of (1) the development of a Python pipeline for the extraction of Available All degrees
models from biological morphological features of 3D, multichannel fluorescence microscopy data of intestinal epithelium
high throughput in different conditions, (2) the descriptive analysis of the epithelial packing
microscopy data (Co- (arrangement of neighbours by distance, cell type, morphology, size) under different conditions, and
supervised with Dr (3) to implement a model that helps predicting the conditions the tissue is experiencing based on the
Joaquina De Navascues, parameters of epithelial arrangement. The project will provide the student the opportunity to participate
from the School of Life in a multidisciplinary project, and to learn methods related to AI, mathematical morphology (for image analysis),
Sciences) network analysis (to develop descriptors) and statistical models in AI (for the predictive models).

Evolved models of Large scale studies that apply next generation sequencing to cancer cohorts, Available All degrees
genomic interactions in are now coming on-line at an ever increasing pace [1]. Knowledge of
AML specific mutations and aberrations is reasonably well defined in most cancers.
However, statistical models and discerning machine learning tools for exploring
the landscape of dependancies between these variables are still lacking.
Of particular interest are patterns of co-mutation and mutual exclusivity.
This project will give opportunity to runn state-of-the-art Bayesian networks learning software [2]
to an AML (acute myeloid leukemia) dataset. Time permiting other approaches for exporing
these can be tested and compare to the contrasting the results. Time and skill permitting there
will be room for methodology development that exploits the structure within the domain. Identifying
alternative datasets is also a direction that can be persued.

[1] Gerstung et al., Nature Genetics 49:332-340 (2017). http://dx.doi.org/10.1038/ng.3756


[2] Gobnilp sofware: https://www.cs.york.ac.uk/aig/sw/gobnilp/

Evolved models of Large scale studies that apply next generation sequencing to cancer cohorts, Available All degrees
genomic interactions in are now coming on-line at an ever increasing pace [1]. Knowledge of
MPN specific mutations and aberrations is reasonably well defined in most cancers.
However, statistical models and discerning machine learning tools for exploring
the landscape of dependancies between these variables are still lacking.
Of particular interest are patterns of co-mutation and mutual exclusivity.
This project will give opportunity to runn state-of-the-art Bayesian networks
learning software [2] to an MPN (myeloproliferative neoplasms) dataset.
Time permiting other approaches for exporing
these can be tested and compare to the contrasting the results. Time and skill permitting there
will be room for methodology development that exploits the structure within the domain. Identifying
alternative datasets is also a direction that can be persued.

[1] Classification and Personalized Prognosis in Myeloproliferative Neoplasms


New England Journal of Medicine, 379:1416-1430
[2] Gobnilp sofware: https://www.cs.york.ac.uk/aig/sw/gobnilp/

Machine learning of gene- One of the unsolved challenges in computational biology is bottom-up prediction of gene regulation Available All degrees
regulation functions (The from interactions of regulatory molecules with the genome. Current high-throughput sequencing
project will be in experiments allow generating large data sets for retrieving the DNA sequence preferences of
collaboration with the regulatory proteins, as well as genomic maps of “epigenetic” chemical modifications and 3D genome
group of Dr Vladimir Teif organization. We developed a MATLAB-based computational framework for predicting the regulatory
at the School of Life protein assembly at genomic regions in chromatin and their effect on gene expression. Your task
Sciences.) will be to tune the parameters of such models using machine learning based on experimental data for stem
cell differentiation. The project can either be within MATLAB, or the function can be called from
Python as to optimise parameters according to experimental data.

A model for predicting One of the unsolved challenges in computational biology is bottom-up prediction of gene regulation from Available All degrees
gene expression in the interactions of regulatory molecules with the genome. Current high-throughput sequencing experiments allow
context of regulatory generating large data sets for retrieving the DNA sequence preferences of regulatory proteins, as well as
protein assembly. (The genomic maps of “epigenetic” chemical modifications and 3D genome organization.
project will be in The biologists have a Fortran and MATLAB-based computational framework for predicting the regulatory protein
collaboration with the assembly at genomic regions in chromatin and their effect on gene expression.
group of Dr Vladimir Teif This project will re-implement the model in either R (preferably) or Python.
at the School of Life Time permitting, the project can then progress into applying machine learning techniques to optimising
Sciences.) the model's parameters to experimental data.

A pipeline for RNAseq The outcome of this project would be an easy to use pipeline implemented in either R or Python, Available All degrees
analysis for differential for the analysis of genome-wide gene expression data (RNAseq). The data will be provided by the
expression of bulk tissue Life Sciences co-supervisor. The primary goal of this analysis would be to identify
based on cell type (Co- differentially expressed genes under several genetic conditions that affect most cell types in
supervised with Dr the tissue of interest. The analysis will include well understood steps such as Gene Set
Joaquín de Navascués, Enrichment Analysis, gene ontology anotation, and combining these with network visualisation of
from the School of Life experimental results based on protein-protein interactions databases (such as STRING). Time
Sciences.) permitting, students will be encouraged to expand the analysis as well as developing novel
algorithms, e.g. using AI approaches and available data identifying the transcriptional
signature of the different cell types in the tissue to identify transcripts more likely
to be differentially expressed in a particular cell type in the experimental conditions.

Visualisation tools for Although Bayesian networks (also known as graphical models) are well regarded disciplined models for Available All degrees
Bayesian networks. exposing the probabilistic dependencies between a number of random variables their usage by non-experts
have been limited. This project will provide a number of graphical tools for visualising BNs produced by
GOBNILP [1]. Implemented in R or Python, the library will provide at a minimum (A) easy control for edge
colours and width and (B) display of heatmaps for each network family. An additional objective for binary
datasets would be (C) the display of logical gates as demonstrated in [2]. At the first instance, static
output will be sought, but linking to interactive network displaying tools will also be an objective if
time permits.

[1] Gobnilp sofware: https://www.cs.york.ac.uk/aig/sw/gobnilp/


[2] The Genomic Landscape and Chronological Reconstruction of Driver Events in Multiple Myeloma

Comprehensive GOBNILP [1] is a machine learning software that maps the Available All degrees
comparison of GOBNILP learning of Bayesian networks (also known as graphical models)
solvers. into an ILP (integer linear programming) optimisation problem
which is then solved by specialist solvers. Currently there
are two implementations: one that is tied to the SCIP optimization
suite and one that depends on Python libraries. Anecdotal evidence
suggest that the Python implementation is less efficient. This project
will extensively test the two implementations on different datasets.
Time permitting the project can then progress to forming hypothesis
that explain any observed discrepancies on performance and their dependence
to the underlying structure of the datasets.
Some simple linux skills will be required in addition to Python.

[1] Gobnilp sofware: https://www.cs.york.ac.uk/aig/sw/gobnilp/


A C-based interface from SWI-Prolog is the most popular implementation of the Prolog programming language Available All degrees
R to SWI-Prolog. which has been extensive used in AI research since the begining of the 1970s.
Although stong in logical reasoning the language is considered specialist.
The ability to call Prolog from other paradigms will be one way to extend its usage.
This project will build a library that will enable Prolog code to be called from within R.
The main vehicle for the interface will be the C-language interface of R and SWI-Prolog.
Ability to program in C is a must. The project will be able to draw ideas from the current
C-based interface that allows R code to be called from within SWI [1].

Real: http://stoics.org.uk/~nicos/sware/real/

Anisi, Hossein Wireless Body Sensor A wireless body sensor network (or simply BSN) is a networked collection of wearable (programmable) sensor nodes that can Available All degrees
Networks for health- communicate among themselves and also with other smart devices and other ambient sensors. The sensor nodes have computation,
monitoring Applications storage, wireless transmission, and sensing capabilities. BSN-based systems can be used to directly monitor several vital signs continuously
and non-invasively. In this project we aim at dealing with one of the following issues: Energy-efficient protocols for body sensor networks,
Frameworks for context-aware body sensor networks, Cloud-assisted body sensor networks, Smart body area networks and Internet of
Things.

Green Data To enable the pervasive deployment of Wireless Sensor Networks (WSNs), the biggest stumbling block is the contradiction between the Available All degrees
Communication in various functionalities demanded by applications and the limited energy supply for sensor nodes. Therefore, in this project, we look at
Wireless Sensor Networks WSNs, mainly from the perspective of green computing in which sensor nodes equipped with certain mechanisms can harvest energy
from surrounding circumstance, or we can rely on dedicated devices to periodically charge the sensor nodes (if they are re-chargeable). In
this project we aim at dealing with one of the following issues: Energy harvesting/charging and power management, Energy-efficient
communication protocol design.

Underwater Acoustic Underwater Acoustic Sensing Networks (UASNs) can be used for smart underwater monitoring and surveillance, off-shore exploration, Available All degrees
Sensor Networks in IoT disaster prevention and recovery, and etc. However, the current technology faces a number of challenges that need to be resolved for
Applications those applications to be cost effective, efficient and practically deployable in large scale in the IoT. The slow propagation speed, low
bandwidth and inherent noise in environment require new or re-design of protocols and algorithms. In this project we aim at dealing with
one of the following issues: Acoustic data collection and processing, Communication architecture for UAWNs, Adhoc network using
autonomous underwater vehicles.

Multimedia-Aware IoT In Internet of things (IoT) computing techniques, multimedia acquisition and communication is very challenging. In such systems, Available All degrees
System Design Image/video data processing, aggregation, and transmission are more complex and resource-consumed. Different and innovative designs
and approaches are necessary for multimedia communication. Moreover, energy-aware and energy-efficient algorithms or approaches
become more essential to prolong the lifetime of the network. In this project we aim at dealing with one of the following issues:
Image/video data aggregation, processing and compression in VSNs, Networking and transmission protocols for VSNs, Unmanned aerial
vehicle (UAV) technology for VSNs.

IoT-enabled With the technical support from Internet of Things (IoT), smart city need to have three features of being instrumented, interconnected Available All degrees
Infrastructures in Smart and intelligent. The explosive growth of Smart City and Internet of Things applications creates many scientific and engineering challenges
Cities that call for ingenious research efforts especially for the development of efficient, scalable, and reliable Smart City based on IoT. New
protocols, architectures, and services are in dire needs to respond for these challenges. In this project we aim at dealing with one of the
following issues: Smart & Efficient Energy Management, Intelligent Traffic Management and Internet of vehicles.

Future Unmanned Aerial Unmanned Aerial Vehicles (UAVs) are equipped with various types of transceivers and sensors, to build seamless communication access Available All degrees
Vehicles Communication platforms. Compared to terrestrial wireless networks, UAVs communication networks have many distinctive features such as high
Networks dynamic network topologies and weakly connected communication links. Therefore, many standards, protocols, and design
methodologies used in terrestrial wireless networks are not directly applicable to airborne communication networks. In this project we
aim at dealing with one of the following issues: QoS and performance for UAVs communication networks, Resource allocation for UAVs,
Dynamic networking and reconstruction of UAV communication network, Bridging and data collection using UAVs

Bartle, Richard A; Lost Cities Player Lost Cities is a two-player card game. The aim of this project is to create a program to play the game. The program must be able to play Available All degrees but
both against a human player and against itself. It can have either a rule-based strategy or a search-based strategy, but it can’t have a particularly the AI and
purely random strategy. The program must play strongly enough to beat a competent human player at least 5% of the time. Games MSc programmes

Player Types Test Some 850,000 people had taken the famous-but-flawed "Bartle Test" when the web site it was on closed down. The objective of this Available All degrees but
project is to create a new Player Types test that better fits the theory. So, that's basically a web interface to a database plus a knowledge particularly the AI and
of Player Types. I can supply the latter... Games MSc programmes

Detective Game The student writes a game in which first a crime is generated procedurally, then the player asks questions to find out whodunnit. It’s open- Available All degrees but
ended how you approach this, but it will probably involve some kind of planning system. You can program it using whatever language you particularly the AI and
choose. You should evaluate how good (or not) the game is by soliciting the view of at least ten players, each of whom has played the Games MSc programmes
completed game at least three times.

Haiku Generator The aim of this project is to create a program that generates plausible English-language Haiku poems. It's fairly easy to write a program Available All degrees but
that generates Haiku (here's one I wrote in Javascript in 2002, which took me about a day: particularly the AI and
http://www.youhaventlived.com/haikumatic/haikum.htm). However, to generate plausible poems requires that the generator has some Games MSc programmes
sense of the meaning of words and how these meanings interact. The core of this project is therefore to do with knowledge
representation (or, conceivably, machine learning) rather than syllable-counting.

Sexism Assessor This is a big data project. The idea is to select papers from a field with which your supervisor is familiar (computer games, AI, in my case) Available All degrees but
and collect as many journal papers, conference papers and book chapters as possible (at least a hundred, preferably more) then run a particularly the AI and
program that you write to extract the references from these papers. For each reference, you determine whether the first author is male, Games MSc programmes
female or indeterminate. You do the same for each paper. You then compare the percentage of papers written by a particular gender
(say, women) with the percentage of citations of articles written by people of the same gender. Is it roughly the same, or do researchers
tend to cite papers disproportionately by author's gender. Do this right and there's a press release in it..!

Tired Kittens Write a single-player Android game about tired kittens, using whatever programming language you choose. You design the game Available All degrees but
yourself. There is great creative freedom here, but please don’t think you can exploit that to avoid doing much programming: the game particularly the AI and
must be uploaded to the Google app store by the time you give your project presentation. Games MSc programmes

Bebbington, David E-band propagation Proposals for spectrum usage include usage of the millimetre waveband, roughly 70-90 GHz, known as ‘E-band’. In this frequency regime, Available MSc Electronic
model the atmosphere and weather has a very strong effect. Unlike the 60GHz band which has extremely string attenuation due to oxygen, the Engineering MSc
E-band lies in an atmospheric window where gaseous attenuation is relatively low, but contributions from water vapour are still Advanced
significant as well as variable. Rain can also result in severe attenuation. The project should aim to quantify these effects (using Communication Systems
statistical data) and to investigate how these would affect system design, taking into account both impairments to transmission as well as
potential mutual interference.

Line of Sight Optical Prior to the advent of optical fibres, the telecommunications backbone in most countries involved microwave radio links. These have Available MSc Electronic
Communication largely become redundant, and current capacity requirements mean that they would never be able to supply the necessary bandwidth. Engineering MSc
Now there are suggestions that the idea of light wave communications through the atmosphere might be revived. One particular Advanced
advantage of this is that light travels 50% faster through the atmosphere than it does through fibre, and in the present day of high speed Communication Systems
financial transactions, communications links with millisecond speed advantages may be worthwhile to specialist operators. There are,
however, downsides to atmospheric communication (besides interception), such as rain, and atmospheric turbulence. This project should
study availability issues, and impairments such as those mentioned, to determine the potential viability of such systems.
Polarimetric and The UK Met Office, and most European meteorological services use rain radar systems operating at C-band (approx. 5.6 GHz), which Available MSc Electronic
radiometric allows sufficiently compact antennas to be used operationally, but suffers from attenuation due to the very rain that is observed. In the Engineering MSc
measurement of UK Met Office’s systems, the transmitter is shut off for short periods to make passive measurements of the thermal radiation emitted by Advanced
precipitation the rain. In principle, these measurements could be augmented by considering the separate contributions from horizontal and vertical Communication Systems
polarizations, which might help in resolving better the depth and intensity of the precipitation. The project involves both simulation and
some data analysis.

Solar/Wind Near Real Renewable energy sources, mainly wind power and solar energy are now contributing significant fractions to the UK grid’s energy supply. Available Any degree
Time Energy Prediction Both these sources are highly variable, however, and there is a need to plan for energy capacity when there is a shortfall in these. To a
considerable extent the two sources are complementary, as it is rarely the case that there are high winds when the weather is fine. This
project aims to look at this issue in greater detail, and to look at how weather prediction data available on-line can be used to plan usage
of resources in advance, and to look retrospectively at how effectively the available resources have been used.

Q-factor maximization in Wireless power transfer systems operating in the MHz frequency bands have been proposed and investigated over a number of years. In Available MSc Electronic
the MHz region these systems, resonant coils are used to set up electromagnetic fields that couple to receiver coils which absorb energy. The distance Engineering MSc
over which this can be done depends in part on the geometry of the coils, but also to a large extent on the quality (Q-)factors of the coils. Advanced
A higher Q-factor means that the coils can store more inductive energy so that a significant portion of it can be absorbed at greater Communication Systems
distance. This project is aimed at determining exactly how the Q-factor can be optimized under the constraints. It will involve some
numerical modelling and calculation.

Aerial Photography There is a very extensive archive of aerial photography over many decades. Typically there are repeat surveys at intervals of some years, Available Any degree
Analysis over which time changes are bound to occur. It would be of interest to be able to survey what changes occur by automated means.
Image processing can be used to this end. This may involve ensuring that images to be compared are properly aligned and to the same
scale, that effective illumination was equalized, after which one would look for changed features. It is envisaged that the project would
make use of available toolkits (such as in matlab, for example) to manipulate images, identify features, and report significant changes in a
way that can be interpreted by a user.

Bourazeri, Katerina A mobile recommender A recommender system provides personalised information/recommendations to users based on their needs and preferences. A Available Any degree
system for healthier recommender system is distinguished in terms of personalisation; it customises its responses to a specific user. This project will focus on
lifestyle the development of a mobile recommender system for healthier lifestyle. It will consider people’s preferences and goals, and also real-
time information (e.g. weather, type of food consumed that specific day, change of plans, etc.) to provide personalised recommendations
to users to help them best achieve their goals towards a healthier way of living.

DNA-Sequencing The project objective is to build a mobile-app game for teaching STEM students about using DNA sequencing to identify diseases. The Available Any degree
Educational Mobile-App game is started with the player receiving information that a number of cases of Beldings disease have been reported in a school. An
Game: “SAVED BY THE information sheet on the disease is given to the players. They are also given some information about each patient. Once players have had
SEQUENCING” a chance to look over the information, they can start to try and make epidemiological links between patients. Then the DNA sequencing
results become available to the player. The player must try to figure out which disease the DNA sequences correspond to.

Designing the New Smart The energy sector faces a rapid transformation lately with significant decentralization and digitization efforts. Increased penetration of Available Any degree
Meter Interfaces distributed energy resources (DER), including PV plants and storage systems, gives an opportunity for disrupting the traditional top-down
philosophy of energy systems. Consumers with domestic generation can be turned into prosumers (i.e. consumers and producers) being
able to generate their own power and support the system in periods of increased stress, further enhancing an energy transition towards
the new paradigm. One of the main challenges lying in this novel concept relates to lack of awareness from the public side regarding the
use of energy. This project will focus on the design of a suitable smart meter interface that could enhance prosumers’ knowledge in this
area.

A Household Energy Serious games are used to train and educate people on science, health care, management and engineering. Training people with serious Available Any degree
Serious Game games is more effective than other educational methods, and users can build knowledge and skills during the gameplay. Energy systems
are a particular area where serious games have been widely deployed to achieve a desired progress or change in player’s behaviour,
while players better understand concepts such as electricity prices and grid sustainability. In this game, players will have to complete
different tasks regarding the energy efficiency of their houses, and they are rewarded for these tasks with points in the game.

A Serious Game for Young Down’s syndrome (also known as trisomy 21) is a genetic disorder caused by the presence of a third copy of chromosome 21 in a baby's Available Any degree
People with Down’s cell. With education and proper care, the quality of life of these people can be improved, and with the right support people with Down's
Syndrome syndrome can have an active role in the community. The objective of this project is to develop a serious game for young people with
Down’s syndrome to engage and encourage them to have a more active involvement in the community. Players will have to complete
mini games related to everyday tasks (e.g. preparing a meal, shopping in the supermarket, paying a bill, buying a ticket to visit a museum
or gallery).

The Effectiveness of Parkinson's disease is a neurodegenerative disorder which leads to progressive deterioration of motor function, including tremor, stiffness, Available Any degree
Exercise in Parkinson's slowness, impaired balance, and later on a shuffling gait. Research has shown that exercise can improve gait, balance, flexibility and
Disease motor coordination. Specific exercises such as walking, cycling, Pilates and yoga can be very beneficial. The objective of this project is to
develop a mobile application which will suggest in real-time the best exercise depending on the specific symptom that people with
Parkinson’s disease have.

Chamberlain, Jon Interactive data Text analytics can provide useful data on human language however it is often difficult for end users to interpret the data. Visualisation Available Not specified
visualisation of text techniques can help present data in a way that supports interpretation and interactivity allowing effective exploration and knowledge
analytics discovery within the data. This project will deliver a novel interactive data visualisation tool for text analytics data derived from social
media conversations, built on open source toolkits such as D3.
https://d3js.org/
For a good example see:
http://www.saifmohammad.com/WebPages/StanceDataset.htm

Visualisation and Anaphora is the linguistic phenomenon where a word or phrase in a text refers back to another entity previously mentioned. For example, Available Not specified
manipulation of in the text “Mary went to the shops and she brought some milk.” the pronoun “she” refers to the previously mentioned entity “Mary”. A
anaphoric chains large corpus of these anaphoric chains was collected using the game-with-a-purpose Phrase Detectives; however, they were collected
from multiple annotators so the chains are noisy. This project will develop a tool to visualise weighted anaphoric links and provide a
method for manipulating the chains.
https://anawiki.essex.ac.uk/phrasedetectives/

Develop a citizen science Games-with-a-purpose (GWAP) has become an increasingly common way to engage participants in science projects and to perform tasks Available Student should ideally
game to label text and/or that would otherwise have to be sent to paid annotators. This project will focus on labelling text documents for linguistic features and/or have a background of
images images for visual features that can be used to train machine learning algorithms. The project will deliver a functioning game prototype, game design
develop the backend database and evaluate the system with user testing.
More details about GWAP for image labelling: https://www.cs.duke.edu/courses/cps296.3/spring07/ieee-gwap.pdf

Dynamic image analysis An example


Image of using
and video stillsGWAP for language
are used labelling: https://www.ijcai.org/Proceedings/15/Papers/596.pdfhttp://ceur-ws.org/Vol-
to build 3-dimensional models of large-scale environments; however, often these images are blurry, Available Student should ideally
and annotation using contain occlusions, or have no features that can be used in the modelling process. By removing these images from the processing pipeline have knowledge of
deep learning the modelling can make more efficient use of resources. Additionally, automatically annotating a complex imageset for objects can prove computer vision
to be difficult for objects with a complex morphology, for example underwater fauna including corals, sponges and algae. Students will
work with the ImageCLEFcoral dataset to develop an annotation technique to attempt to beat the existing benchmarks of 2019 with the
goal of competing in the task for 2020.

https://www.imageclef.org/2019/coral
http://ceur-ws.org/Vol-2380/paper_200.pdf
Automated search query Finding the right information at the right time is a constant challenge. Sometimes, a few keywords in a search box is good enough. But Available Student should ideally
generation (co-supervised there are times when a more rigorous, precise approach is needed. For example, 2dSearch is a radical alternative to conventional have a text
by Prof Russell-Rose) 'advanced search'. Instead of entering Boolean strings into one-dimensional search boxes, queries are formulated by combining objects analytics/information
on a two-dimensional canvas. This eliminates syntax errors, makes the query semantics more transparent, and offers a more effective retreival background
way to optimize, save and share search strategies.

However, crafting effective search strategies remains a skilled task, requiring expertise and judgement. The goal of the project is to
investigate ways in which automated support might be provided. For example, given a job spec, can we automatically generate an
effective search strategy that finds social profiles of suitable candidates? What kinds of terms should be extracted, and how should they
be structured? How might we evaluate such a tool? The deliverable should be a working prototype and an evaluation of its performance
(test data can be provided).

https://www.2dsearch.com/

https://www.researchgate.net/publication/326669133_Preprint_submitted_to_Journal_of_Information_Processing_and_Management
_Information_Retrieval_in_the_Workplace_A_Comparison_of_Professional_Search_Practices

An iPlayer for search Not all search tasks start from a blank canvas. In fact, many search problems are variations on an existing theme. In healthcare, for Available Student should ideally
strategies (co-supervised example, information specialists draw on repositories such as the ISSG Search Filters as a source of best practice and predefined have a text
by Prof Russell-Rose) templates (or ‘search strategies’). In cases such as these, 2dSearch has the ability to parse such templates and render them directly on analytics/information
the canvas. But by rendering them instantaneously we miss a crucial opportunity to help the user understand and interact with that retreival background
content.

The goal of this project is to develop an ‘iPlayer’ for search strategies, allowing them to be downloaded and ‘played’ on demand as
executable specifications that ‘build’ in front of the user, from a blank canvas to a completed search strategy. Such a player could provide
vital insight into the efficacy and semantics of legacy search strategies, showing in real time how they were constructed, allowing the
user to pause and interrogate them at various stages, and, crucially, highlighting the various ways in which they could be optimised,
extended, or re-used. The deliverable should be a working prototype and some evaluation of its performance (test data can be provided).

For examples of healthcare search strategies see:


https://medinform.jmir.org/2017/4/e33/
Mechanical interventions Biological recruitment to underwater structures that have been deployed for longer than a few days may impede mechanical operation Available Student should ideally
to prevent biofouling of of moving parts and the effectiveness of operation. Typical solutions use highly toxic paints to prevent biofouling; however, knowledge of have an engineering
underwater machinery biological succession (the order of what marine species start growing on structures) with innovative self-cleaning parts could reduce the background
(industry project) ecological footprint of such machinery. This project is in collaboration with an industry partner who would support the specification of the
machinery, provide details of marine areas of interest and support the deployment and recovery of test equipment.

Underwater search and This project would aim to improve the efficiency of the search and recovery process used in surveying underwater areas for metallic Available Student should ideally
recovery using 3D objects. Our premise is that 3D mapping can be done rapidly in areas where metal objects are detected from the surface by adapting an have an engineering
mapping on an ROV ROV fitted with metal detector. In this way an operator can find objects and survey areas quickly, leading to improved decision making for and/or computer vision
(industry project) recovery. This project is in collaboration with an industry partner who would provide the parts and support the operational testing of the background
device either in a pool or shallow water area.

Chowdhury, Anirban Assist-as-needed control Robotic exoskeletons are the next generation of robots ready to revolutionize the field of augmented human capacity and making Available Any degree
for robotic hand- enabling technologies for the disabled population. A major problem in this area is to optimally challenge the users to avoid over-
exoskeleton dependency on the exoskeleton assistance. The purpose of building an assist-as-needed controller is to solve this problem by adjusting the
difficulty level of an exoskeleton assisted physical practice task intended to facilitate motor learning. The challenge is to come up with an
optimal strategy to tune the parameters of an impedance controller by getting biofeedback from the users about their motor ability.

Robot motion planning Humans are mostly dependent on vision for gross manipulation of their limbs and in robotics people often try to incorporate such ideas in Available Any degree
using visual generalized robot manipulation. A new development in robot motion planning is the use of visual generalized co-ordinates (VGC) in order to use vision
co-ordinates. to identify its configuration and environment. This method is an alternative to the traditional C-space approach of robot motion planning.
VGC leads to the creation of a Visual Roadmap (VRM). It deals with solving inverse kinematics visually using an interpolation of nearby
images from an image space of different poses of the robot. This solution facilitates the planning path for manipulators with unknown link
geometries.

Redundancy resolution of Solving the inverse kinematics of hyper-redundant robot manipulators is a major challenge in robotics. Existing solutions are Available Any degree
hyper-redundant robot computationally complex and difficult to implement in a real-time setting. While most of the robotic applications demand real-time
manipulation. solutions, building cost-efficient mechanisms of redundancy resolution are the need of the hour. The project will investigate the
geometric approach for such a solution. The first step would be to make a solution for hyper-redundant 2-D manipulators and then
extending it to 3D. The practical validation of the geometric algorithm will also be involved in the study to show its efficiency in a real-
world setting.

Inter-subject transfer Conventional brain-computer interfaces need recalibration every time when a user sits for using it. This the process is very time Available Any degree
learning for MEG based consuming and becomes more cumbersome when the BCIs are intended for neurorehabilitation where the stroke patients are involved.
brain-computer Inter-subject transfer learning can reduce the need for recalibration by using past data from other subjects and hence previously explored
interfaces. for EEG based BCI. The challenge is to implement this strategy for MEG based BCI systems which are very promising for designing
neurorehabilitation paradigms as they have more spatio-temporal resolution. The project investigates different machine learning
algorithms for inter-subject transfer learning relevant to MEG BCI.

Human-robot co- Human-robot co-operation is an area where robot and human comes together rather than coming against each other in order to Available Any degree
operation using brain- enhance productivity in an industrial or social scenario. It is very essential for robots and humans to understand each other in order to co-
computer interface based operate in a synergistic way. Unlike the robots humans are subjected to change in their emotional states which affect their performance.
mental state observation. Here the idea is to use passive brain-computer interfaces to convey the mental states of a human to the robot so that the robot can
change its control parameters for better interaction with the human counter-part in a human-robot co-manipulation scenario.

Reconfigurable modular Navigation in an unstructured environment is a major challenge in robotics. Different locomotion strategies have their own advantages Available Any degree
robots for navigation in and disadvantages and they are suitable for different terrains. Building a mobile robot adapting to different environmental conditions
an unstructured needs modularity in the design. The robot modules can be reconfigured to suit different terrains to adopt different locomotion strategies
environment. just like the “Transformers”. The challenge is to develop various locomotion strategies and ways to control them in an optimal way to
facilitate the terrain adaptability for mobile robots. The project takes a mechatronics approach of innovation through the integration of
mechanism design, embedded systems, and algorithms.

Citi, Luca Taking part to a Kaggle Kaggle is a platform for predictive modelling competitions. The top players in the field of data science (universities and companies) Available MSc Advanced Computer
(or similar) Machine organize competitions, provide the data, and score the results. The leading experts in Machine Learning compete against each other to Science, MSc Artificial
Learning competition squeeze as much information as possible from those data and be at the top of the competition chart. Do you like the idea? Find a Intelligence, MSc Big
challenge, make an appointment to discuss it with me and we'll give it a shot. Data and Text Analytics,
MSc Intelligent Systems
and Robotics, MSc Data
Science, MSc Computer
Games
Student’s own It is open to students to suggest topics for projects in the areas of data science, robotics, game AI, text analytics, using Bayesian Deep Available MSc Advanced Computer
suggestions for projects Learning ( https://arxiv.org/pdf/1604.01662.pdf ). Please make an appointment to discuss possibilities (please bear in mind this needs to Science, MSc Artificial
using Bayesian Deep be done in a timely manner). Intelligence, MSc Big
Learning Data and Text Analytics,
MSc Intelligent Systems
and Robotics, MSc Data
Science, MSc Computer
Games

AI agent for general video This project involves the design and implementation of an AI agent able to play video games within the General Video Game AI Available MSc Advanced Computer
games framework (GVGAI www.gvgai.net). The student can design the AI taking inspiration from existing approaches such as Monte Carlo Tree Science, MSc Artificial
Search (MCTS), Genetic Algorithms (GA), Deep Learning, and others. Intelligence, MSc Big
Data and Text Analytics,
MSc Intelligent Systems
and Robotics, MSc Data
Science, MSc Computer
Games

Own project in Machine It is up to the student to propose a topic which builds on the following literature: Available MSc Advanced Computer
Learning based on list of http://proceedings.mlr.press/v97/liu19b/liu19b.pdf Science, MSc Artificial
references https://arxiv.org/abs/1808.02610 Intelligence, MSc Big
https://arxiv.org/abs/1802.09098 Data and Text Analytics,
https://arxiv.org/abs/1802.07814 MSc Intelligent Systems
https://arxiv.org/abs/1602.04938 and Robotics, MSc Data
Science, MSc Computer
Games

Efficient parametrisation Markov-Chain Monte Carlo techniques are often applied to solve integration and optimisation problems in large dimensional spaces [1]. Available MSc Advanced Computer
of state space models for These problems play a fundamental role in machine learning, and in particular in Bayesian inference and learning. Science, MSc Artificial
Bayesian inference in State-space models are a popular framework to model time series (physiological data, financial markets, etc) but unfortunately the Intelligence, MSc Big
Markov-Chain Monte parametrisation of state-space models within standard MCMC tools is far from trivial (multiple modes, stability, etc). This project will Data and Text Analytics,
Carlo samplers investigate novel techniques to tackle these issues and simplify the parametrisation of state-space models for use in MCMC tools and MSc Intelligent Systems
simplify their application to time series modelling. and Robotics, MSc Data
Science, MSc Computer
This project is better suited to students with strong mathematical background and familiarity with state space models (e.g. a first degree Games
in control theory or similar).
[1] https://doi.org/10.1023/A:1020281327116

Efficient parametrisation Markov-Chain Monte Carlo techniques are often applied to solve integration and optimisation problems in large dimensional spaces [1]. Available MSc Advanced Computer
of orthogonal matrices These problems play a fundamental role in machine learning, and in particular in Bayesian inference and learning. Science, MSc Artificial
for Bayesian inference in Statistical and ML models parametrised in terms of orthogonal matrices are ubiquitous but unfortunately the parametrisation of Intelligence, MSc Big
Markov-Chain Monte orthogonal matrices within standard MCMC tools is far from trivial [2,3,4]. In this project the student will refine and thoroughly test a Data and Text Analytics,
Carlo samplers novel algorithm which greatly simplifies the parametrisation of orthogonal matrices. MSc Intelligent Systems
and Robotics, MSc Data
This project is better suited to students with strong backgorund in linaera algebra and statistics. Science, MSc Computer
[1] https://doi.org/10.1023/A:1020281327116 Games
[2] https://discourse.mc-stan.org/t/efficient-orthogonal-matrix-parameterization/8685/7
[3] https://discourse.mc-stan.org/t/new-transform-for-orthonormal-matrices-in-stan/2981
[4] https://arxiv.org/pdf/1710.09443.pdf

Clark, Adrian Making augmented Augmented Reality (AR) is widely regarded as a key technology in the next few years. An important factor in producing realistic-looking Available Any degree
reality match the real AR is matching the virtual camera used in graphical components to match the one used in the real world. This project will devise practical
world ways by which this can be done, either by calibration in advance or dynamically as the AR system runs.

Tracking football players Knowing the locations of players in a football match lets coaches use VR reconstructions as a training aid. However, tracking players fully Available Any degree
during matches automatically is very difficult to do. This project will develop a tool that automates tracking where possible and uses human interaction
when tracking is lost.

Annotation of 3D models The annotation of training data for machine learning systems is a major headache. Tools are starting to appear for annotating images Available Any degree
rapidly and easily, both from academic research and industry. However, there have been very few attempts to produce tools for
annotating 3D content, such as the point clouds produced by structure-from-motion systems. The purpose of this project is to develop
such a tool and to demonstrate it on coral reef imagery and 3D models that have been captured in CSEE as part of a major project on reef
reconstruction.

A ‘virtual trainer’ serious People who have suffered from fractured limbs or surgery such as hip replacements are usually given exercises to do at home to aid their Available Any degree
game recovery. However, most people either don’t do them or do them ineffectively. This project will use OpenPose (see
https://www.youtube.com/watch?v=pW6nZXeWlGM) to identify the skeleton of a person as they carry out their exercise and encourage
them to match the pose of a ‘virtual trainer’. The project will be demonstrated in a state-of-the-art VR/AR facility in CSEE.

Tracking skeletons The OpenPose toolkit (see https://www.youtube.com/watch?v=pW6nZXeWlGM for what it can achieve) is able to fit skeletons to people Available Any degree
in videos at frame rates. However, identifier that references a particular skeleton can change from frame to frame as the participants
move. The purpose of this project is to devise a way to track skeletons through a video, and the intention is to demonstrate it on both
video sequences and live imagery.

Monitoring people for We are developing autonomous systems that monitor elderly and vulnerable people, allowing them to live in familiar surroundings rather Available Any degree
their activity than in care homes as this greatly improves their quality of life. An important aspect of this is to identifying how frequently people move is
this is a major factor in maintaining their overall health. OpenPose (see https://www.youtube.com/watch?v=pW6nZXeWlGM for
example) gives a way of doing this, and the purpose of this project is to develop a prototype system that monitors for how often and how
much people move and compare the results with fitbit-type devices.

Daly, Ian Brain stimulation and Our brains may be thought of as networks of interacting regions. The connections between these regions may be explored by stimulating Available MSc Computer Science;
computational modelling the brain. Using a custom built robot to magnetically stimulate the brain this project will explore connections between brain regions and Prerequisites: Useful:
of brain connectivity build computational models of how different regions of the brain interact with one another. Computational modelling techniques will be Signal processing,
developed / adapted to model how different regions of the brain interact with one another as we attempt to move. Machine learning, any
experience with /
knowledge of EEG or BCI

Building a fast Brain- Brain-computer interfaces (BCIs) allow their users to control a computer directly with their brain activity. This project will develop a new Available MSc Computer Science;
computer interface type of BCI based on coded visual evoked potentials to allow very fast and accurate decoding of intention from the brain. Existing state-of- Prerequisites: Essential:
the-art tools for BCI development (such as openVibe, BCI200 etc) may be used to develop the BCI. Additionally, a new visual interface will Signal processing. Useful:
be developed. The project will include working with state-of-the-art signal processing and machine learning methods as well as direct Machine learning, any
experimentation with EEG systems in the Essex BCI lab. experience with /
knowledge of EEG or BCI
Subliminal Brain- Brain-computer interfaces (BCIs) are often controlled by the event related potential (ERP) (a particular pattern of neural activity). The Available MSc Computer Science;
computer interfacing speed at which the ERP can be detect directly effects the communcation speed of the BCI. This project will explore and develop a novel, Prerequisites: Essential:
fast, type of BCI based on subliminal stimuli (stimuli presented too fast for consious perception). The student will construct and evaluate a Signal processing. Useful:
BCI using subliminal stimuli by first developing and validating a visual interface capable of presenting subliminal stimuli, followed by Machine learning, any
development of other parts of the BCI pipeline (signal processing and state-of-the-art machine learning methods). Finally, the student will experience with /
test the system with EEG experimentation. knowledge of EEG or BCI

Movement trajectory Some research has suggested it may be possible to identify movement trajectories from EEG via simple low-pass filtering and correlative Available MSc Computer Science;
tracking from EEG analysis (see Bradbury, 2010, and Ortner, 2013). If correct, this could be used for more accurate limb position tracking and development Prerequisites: Essential:
of movement trajectory based BCIs. This project will seek to validate this line of research via careful controlled experimentation. The Signal processing. Useful:
student will develop the methods (signal processing and machine learning) to explore this problem. They will also design and run Machine learning, any
experiments. experience with /
knowledge of EEG or BCI

Machine learning to Biomedical signals such as electrocardiogram (ECG, heart rate) or electroencephalogram (EEG, brain signals) may be used to diagnose a Available MSc Computer Science;
classify biomedical data wide range of conditions from hearth disease through to alcoloism. Machine learning tools may be used to automatically diagnose Prerequisites: Machine
conditions from these datasets. However, the large number of different machine learning tools available, and their complexity, means learning. Useful: Signal
that specialist skills are needed to apply these tools to biomedical datasets. In this project the student will develop a platform to apply a processing.
range of state-of-the-art machine learning / deep learning tools to biomedical datasets, with the aim of diagnosing health concerns. As a
starting point, data from the kaggle data hosting platform will be used.

Efficient emotion Emotion is a core, yet often overlooked, element of providing care to patients and the elderly. Rapid changes in emotion can precede falls Available MSc Computer Science;
detection algorithms or other health problems. In this project the student will develop efficient algorithms for rapidly assessing the current emotion of an Prerequisites: Some
individual based on signals recorded from a wearable device. The student will need to develop signal processing and classification signal processing
algorithms to monitor changing phsyiologicaly signals (heart rate, blood pressure etc.) to assess a persons current emotions. A wearable experience.
device will be provided that is able to measure heart rate and other physiological signals.

Extending multi-voxel Multi-voxel pattern analysis is a machine learning method that allows identification of patterns of voxels within a functional magnetic Available MSc Computer Science;
pattern analysis to joint resonance image (fMRI) that correlate with one or more stimuli. It is a powerful technique that has been used to identify patterns of Prerequisites: Machine
EEG-fMRI recordings activity related to a wide range of cognitive processes, from focus on specific nouns through to emotional responses to stimuli. Joint EEG- Learning.
fMRI recordings a a new imaging modality that allow epxerimenters to record simultnaoeus measures of the electrical and
heamodynamic activity in the brain. By applying MVPA to such joint recordings it may be possible to identify patterns of activity in the
brain with greater precision than previously possible. In this project the student will extend the current MVPA algorithm to such multi-
modal datasets.

Student proposed project I am happy to discuss with students their own ideas for MSc projects in the areas of brain-computer interfacing or neural engineering. Available MSc Computer Science
in Brain-computer
interfacing / neural
engineering

Daly, Ian and Soni, Machine learning based Emotion is a core, yet often overlooked, element of providing care to patients and the elderly. Rapid changes in emotion can precede falls Available MSc Computer Science,
Aakash emotion detection using or other health problems. In this project the student will develop efficient algorithms for rapidly assessing the current emotion of an MSc Intelligent Systems
PPG and ECG sensors. individual based on signals recorded from ECG and PPG sensors. The student will need to develop signal processing and classification and Robotics;
algorithms to monitor changing phsyiologicaly signals (heart rate, blood pressure etc.) to assess a persons current emotions. Moreover, Prerequisites: Essential:
the hardware platform will be developed that is able to measure heart rate and other physiological signals. Finally, a comparative analysis Signal processing,
will be carried out to examine the performance of PPG and ECG sensors. Machine learning, C and
Python programming.
Useful: Any experience
with / knowledge of
ECG/PPG and embedded
systems

Daly, Ian and Soni, Fuzzy logic based emotion Emotion is a core, yet often overlooked, element of providing care to patients and the elderly. Rapid changes in emotion can precede falls Available MSc Computer Science,
Aakash detection using PPG and or other health problems. In this project the student will develop efficient algorithms for rapidly assessing the current emotion of an MSc Intelligent Systems
ECG sensors. individual based on signals recorded from ECG and PPG sensors. The student will need to develop signal processing and classification and Robotics;
algorithms to monitor changing phsyiologicaly signals (heart rate, blood pressure etc.) to assess a persons current emotions. Moreover, Prerequisites: Essential:
the hardware platform will be developed that is able to measure heart rate and other physiological signals. Finally, a comparative analysis Signal processing, Fuzzy
will be carried out to examine the performance of PPG and ECG sensors. logic, C programming.
Useful: Any experience
with / knowledge of
ECG/PPG and embedded
systems

Daly, Ian and Soni, Assessment using ECG Biomedical signals from electrocardiogram (ECG) and photoplethysmogram (PPG) sensors may be used to diagnose a wide range of Available MSc Computer Science,
Aakash and PPG sensors for conditions from cardovascular disease through to cardovascular fitness. Machine learning tools may be used to automatically diagnose MSc Intelligent Systems
emotion detection. and recognise conditions. The large number of different machine learning tools available, and their complexity, means that specialist skills and Robotics;
are needed to apply these tools to biomedical signals and their datasets. In this project the student will develop a platform to apply a Prerequisites: Essential:
range of state-of-the-art machine learning tools to biomedical datasets, with the aim of generating health report based on measured Signal processing,
health variables such as hear rate, HRV, blood oxygen level, blood pressure and so on. As a starting point, a dataset will be created and Machine learning, C and
public datasets will be invesitgated. Python programming.
Useful: Any experience
with / knowledge of
ECG/PPG and embedded
systems

Doctor, Faiyaz Agent Based Models for The aviation industry is a significant contributor to the burning of fossil fuels which is a leading cause of climate change. Future Available Any degree
Future Airport electrification of air transport infrastructure has emerged as a key driver to facilitate transitions to more sustainable future air travel.
Infrastructure Physical infrastructure challenges relate primarily to how best to safely and efficiently integrate electric aircraft into legacy airport
Requirements Simulation systems and supply chains. These may include, but are not limited to, issues around the optimal nature, design and location of in-house
charging facilities, use of smart grids, requirements of specialist transport and lifting equipment for ‘drop-in’ batteries, as well as
certification of safe connections between aircraft and charging infrastructure. There are also questions in terms of electrical power and
distribution requirements as well as related issues of resilience, contingency planning and cyber-security. This project should design and
prototype an agent-based modelling to develop virtual physical airport infrastructure models for representing complex resource
allocation and simulating operational scenarios around electrification of airport infrastructure. Computational nature inspired
optimisation techniques should also be investigated for integration with the simulation model to generate and evaluate optimal solutions
to complex resource allocation and operations challenges.
Development of Smart VR Universities have historically been involved in the process of knowledge creation, primarily through research, knowledge transfer (KT), Available Any degree
Tools for Innovation and teaching. There is a need for universities to develop better digital tools to facilitate innovation and entrepreneurship training through
Entrepreneurship which both staff and students can better exploit intellectual property created through research, KT and educational activities. The use of
Training existing popular fantasy and science-fiction can be harnessed to help facilitate the process of creativity and innovation. Diegetic
Innovation Templating (DiT) is a methodology in which an artefact is extracted from a fantasy or science-fiction story that captures an
innovative concept. The artefact can then be abstracted into an innovative product or process design (Zheng and Callaghan, 2018). DiT
could therefore be used as part of a training methodology for teaching creative design, product development and entrepreneurship for
the engineering and technology sector. The use of Virtual Reality (VR) is increasingly being used to provide an effective tool for engaging
staff, students and industrial professionals through creating tailored and immersive experiences that can be designed to convey
information in an intuitive and interactive manner. The purpose of this project is to design and develop a VR based collaborative training
tool for staff that incorporates the DiT methodology for designing and delivering innovation and entrepreneurship training programs. The
VR training content should be rendered and displayed based on the interaction of user with the tools the interactions and information
seeking behaviour of users might be uncertain and vaguely defined. Fuzzy logic can provide a robust methodology for handling and
reasoning about imprecise information and can be used to control how interactive components behave and how content is displayed to
the user.

Development of Virtual The iSpace at the university of Essex is a test bed for smart homes technologies and the Internet of Things. The iSpace is a fully- Available Any degree
iSpace Software functioning, self-contained flat comprising of heterogeneous smart devices, actuators and sensors. These artefacts communicate over
Emulation wireless PAN, HAN protocols operating over the Physical and MAC Layers of an Internet of Things stack. Other connected devices
communicate over propriety networks and WiFi. There is a need to complement the physical infrastructure of the iSpace with an identical
simulated virtual representation which can be used to both interact and control the physical sensors and actuators as well as virtually
simulate operational configurations and scenarios. The aim of the project is to design and develop a prototype functioning virtual
software emulation of the iSpace that can be used by staff and students to control configure and emulate the real environment as well as
provide a secure sandbox to evaluate intelligent agents for interactive or automatic control.

AR-supported remote The project proposal is about one use case of Augmented worker BT research project where worker could be desk or field agents able to Available Any degree
assistance (industrial consume some information on the top of what they already have access to thanks to Augmented Reality. The use case proposed in the
project with BT, co- proposal here works with nReal-based AR application –some very initial POC developed in BT will be available- where an engineer is
supervised by Dr. Anne requesting a list of contact able to assist on a given list of criteria.
Liret) There is a mix of development of the criteria selection app, and the research bit is the “recommender system” which is going to for
instance use a Machine learning model or a classifier coupled to similarity computations to extract who was able to complete similar type
of work (work having similar features as the criteria selected by the user) and recommend the most suitable (there can be a fitness
function that drives to find the engineer having completing work matching the highest number of criteria, or matching a set of weighted
criteria with the highest score, or having complete the work with the higher effectiveness compared to the standard time expected for
the work).

Skills requirement
Programming in java, javascript, eventually with Android environment,
Some knowledge in classification techniques (tf-idf, doc2vec) and python will really help too.

Gan, J Machine Learning for Image classification is a challenging problem. This project aims to implement and evaluate machine learning methods, including deep Available MSc Advanced Computer
Image Classification learning, for image classification. Your task is to implement several machine learning methods and evaluate their image classification Science, MSc Artificial
performance using the ImageNet database (http://image-net.org/index). Third party tools/libraries can be used for implementing Intelligence
machine learning methods. You are expected to find a good machine learning method for image classification and evaluate it using image
classification competition data (e.g., http://image-net.org/challenges/LSVRC /2017/). You can find more details about machine learning
for image classification from the following papers:
Alex Krizhevsky, Ilya Sutskever, and Geoff Hinton, “ImageNet classification with deep convolutional neural networks,” Advances in Neural
Information Processing Systems, vol. 25, pp. 1006-1114, 2012.
Waseem Rawat and Zenghui Wang, “Deep convolutional neural networks for image classification: A comprehensive review,” Neural
Computation, vol. 29, pp. 2352-2449, 2017.

ImageNet Object This project aims to develop, implement and evaluate methods for detecting and identifying objects from images in the ImageNet Available MSc Advanced Computer
Identification Challenge database. An image may contain several different objects. Your task is to identify and label objects from images of over 1000 categories. Science, MSc Artificial
Detailed description of the task and examples of images can be found at https://www.kaggle.com/c/imagenet-object-localization- Intelligence
challenge. You are allowed to use third party tools/libraries for implementing conventional computer vision methods and machine
learning methods. You are expected to evaluate several methods so as to choose the best for preparing a possible entry to the ImageNet
Object Localisation Challenge.

Developing an app for Affective computing makes it possible for a machine to interpret the emotional state of human beings and adapt its behaviour for more Available MSc Advanced Computer
Emotion Recognition natural human-machine interaction. This project aims to develop an app for emotion recognition from facial images. Different computer Science, MSc Artificial
from Facial Images vision and machine learning methods for emotion recognition from facial images will be investigated using a large database of facial Intelligence
images. The best methods found will be implemented in a smartphone app with a user-friendly GUI. You are allowed to use third party
tools/libraries for implementing conventional computer vision methods and machine learning methods.

An Assistive System for Kaggle (https://www.kaggle.com/) is a company that operates online machine learning /data science competitions. The basic idea is very Available MSc Advanced Computer
Preparing Entries to simple. You are supplied with a training set and a test set. However, the test set does not include the values of the variable to be Science, MSc Artificial
Kaggle Machine Learning predicted. Once training is complete, you run your solution on the test set and then upload your predictions to Kaggle who will give you an Intelligence
Competitions evaluation score. This project aims to develop an assistive system with a GUI for preparing entries to Kaggle competitions. The system
should have the following basic functions: importing/loading a dataset and dividing it into training and validation subsets in a random
manner, pre-processing the data and extracting features using different methods depending on the problem domain, classifying the
extracted features using at least two classifiers, evaluating the performances of different methods through statistical significance test, a
GUI for model/parameter settings, results visualising and saving, etc.. Third party tools/libraries can be used for implementing machine
learning methods.

Developing an AI Player I would like to discuss your original ideas about a board game and advise you on possible AI algorithms to be considered for game playing Available MSc Advanced Computer
for a Board Game (of and requirements of the AI player. Science, MSc Artificial
Your Choice) Intelligence

Developing an AI app (for I would like to discuss your original ideas and advise you on both the problem domain and related AI techniques. Available MSc Advanced Computer
something of your choice) Science, MSc Artificial
Intelligence

Gardner, M R; An Online Innovation Lab This project concerns building a virtual-reality prototype of an innovation lab. Existing research shows that people are more creative Available MSc. Computer Science,
(in conjunction with Vic when holding conversations about solving problems or generating new products, when they are in informal spaces such as cafeterias, bars MSc. Advanced Web
Callaghan ) or even bespoke playful/gaming environments. This project aims to create a simple prototype of such a playful ‘ideation’ space, in virtual Engineering
reality. The project will be supported by the Creative Science Foundation (CSf) - www.creative-science.org) who had been considering
using Amazon Lumberyard as the development environment and operational platform (https://aws.amazon.com/lumberyard/), since it
provides a cloud platform that reaches into countries like China (who traditionally have firewalls in place) and provides a comprehensive
but simple to use set of tools to construct VR worlds. That said, if a student has a preference for another tool (eg existing knowledge of
another environment such as https://unity.com/) CSf are open to the use of other VR tools. More information can be found at
https://www.creative-science.org/activities/ourhex/ or by contacting either Michael Gardner (mgardner@essex.ac.uk) or Victor
Callaghan (vic@essex.ac.uk)
SolveMy Problem (in In simple terms, this project concerns the development of a bespoke website (or possibly an App), together with a supporting online set of Available MSc. Computer Science,
conjunction with Vic tools, that bring together people with know-how, with those who need it. There are a number of websites which, superficially, have MSc. Advanced Web
Callaghan) similar goals (eg https://www.kolabtree.com/how-it-works, https://www.upwork.com/ etc) but the aim of this project goes further in Engineering
that it seeks to lever the latent capability in existing networks of researchers and academics that coalesce around specialist conferences
or organisations as well as addressing the broader eco-system of issues that any given expertise requirement operates within. For the
purposes of this Masters project, the student will be supported by two international organisations (iLRN - https://immersivelrn.org & CSf
– http://creative-science.org), creating a demonstrator that works for these organisations and for their knowledge base. For more
information on this project look at the iLRN & CSf websites (see above) or for more general reading about the wider issues involved look
at http://www.journaliji.org/index.php/iji/article/view/79/_2 or email either Michael Gardner (mgardner@essex.ac.uk) or Victor
Callaghan (vic@essex.ac.uk)

A web VR visualisation BIM (Building Information Modeling) is an intelligent 3D model-based process that gives architecture, engineering, and construction Available MSc. Computer Science,
tool for BIM models (AEC) professionals the insight and tools to more efficiently plan, design, construct, and manage buildings and infrastructure. This project MSc. Advanced Web
(industrial project with seeks to create a WebGL VR application to automatically upload a BIM model, to allow visulaisation of infrastructural installations. This is Engineering
BT, co-supervised by Dr. a collaborative project with BT, and as such, the content will be designed for field engineers tasks.
Anasol Pena-Rios) Deliverables
• Build a VR application able to identify, import and integrate Building Information Models.
• Full documentation including technical documentation and user manuals.
Required technical skills and knowledge
• Programming skills in Phyton and game programming (Unity, C#)
• Knowledge of 3D modelling (desirable, not compulsory)

A web 360 VR simulator Talent attraction is an important activity at BT. We currently use a mobile VR app to showcase everyday working scenarios for specific Available MSc. Computer Science,
(industrial project with roles within BT using 360 videos. This project seeks to recreate the same experience using a more accesible medium as a browser. This MSc. Advanced Web
BT, co-supervised by Dr. should include elements of gamification to enhance user experience, and ensure that users look at all the information available within the Engineering
Anasol Pena-Rios) Web VR experience. This is a collaborative project with BT, and as such, the content will be designed for field engineers tasks and training.
Deliverables
• A customisable gamified web VR app for talent attraction purposes in BT.
• Full documentation including technical documentation and user manuals.
Required technical skills and knowledge
• Good programming skills in game programming
• Knowledge of web development
• Knowledge of React 360 (desirable)

Optimisation of Virtual Creating virtual representations of complex spaces can generate multiple objects at the same time, which can slow down the simulation Available MSc. Computer Science,
Objects in Simulated generating a bad user experience. This project seeks to identify which objects render in base a certain conditions such as visibility from MSc. Advanced Web
Environments (industrial the user to the object(s). An alternative could be to reduce resolution depending on the distance between the user and the object. This Engineering
project with BT, co- can be very useful in complex visualisations such as BIM models. BIM (Building Information Modeling) is an intelligent 3D model-based
supervised by Dr. Anasol process that gives architecture, engineering, and construction (AEC) professionals the insight and tools to more efficiently plan, design,
Pena-Rios) construct, and manage buildings and infrastructure. This is a collaborative project with BT, and as such, the content will be designed for
field engineers tasks and training.
Deliverables
• A Web VR application to vsualise optimised complex representations
• Full documentation including technical documentation and user manuals.
Required technical skills and knowledge
• Programming skills in Phyton and game programming (Unity, C#)
• Knowledge of 3D modelling (desirable, not compulsory)

Gu, Dongbing Object Instance Object instance segmentation is to detect objects from an image and generate a high quality segmentation mask for each instance. This Available All degrees
Segmentation and 3D can be done by training a deep learning network with labelled datasets. An successful example is the Mask RCNN
Point Cloud Construction (https://arxiv.org/abs/1703.06870). Based on object instance segmentation, this project is to detect moving objects from a video and
from video construct an 3D point cloud from each detected instance. Once object instances are detected, their depths should be estimated
simultaneously from another deep neural network. the depth networks can be trained independently with unsupervised
datasets(https://arxiv.org/abs/1603.04992). The 3D point cloud can be reconstructed from an instance mask and its corresponding
depth map.

Estimating depth from Currently the depth map can be estimated from monocular images or video clips by using deep learning networks. In some cases, the Available All degrees
monocular images or deep learning network can be trained using stereo images or videos to learn the absolute scale, and the depth map can be inferred by just
video clips feeding monocular images or video clips to the networks. This project is to explore the advantage of using Siamese networks for depth
estimation. A reference can be found in this link https://arxiv.org/abs/1905.00401. In addition to the depth estimation, the camera pose
should also be estimated simultaneously by adding an additional deep network. An example of simultaneously estimating the depth and
camera pose using unlabelled datasets can be found here (https://arxiv.org/abs/1709.06841).

6D Object Pose One of the popular methods for robotic manipulation is to estimate the 6D object pose for an object first, then perform the manipulation Available All degrees
Estimation using RGB-D accordingly. The RGB-D camera provides two sources of information for such a purpose. The estimation of 6D object pose can be
camera for Robotic implemented by using deep learning networks. The key problem is to identify what kind of system architecture is appropriate and what
Manipulation kind of datasets are available for training the system. A reference can be found from this link: https://arxiv.org/abs/1901.04780.

Machine Learning The project aims to develop machine learning algorithms for condition monitoring on Heating, Ventilation, and Air Conditioning (HVAC)
Algorithms of Condition Systems. HVAC turbines and its control systems should be monitored by vibration and electric current sensors. The air quality of
Monitoring on HVAC environment should be monitored by temperature, humanity, pressure, and CO2 sensors. Given these sensor readings, how to find out the
Systems machinery fault, and control the air quality is a challenging task. The project investigates how to use machine learning algorithms to solve
the problems. An additional feature is to coordinate multiple HVAC Systems to improve the air quality of environment. The project is
linked to a local company Flakt Woods.

Visual guided UAV UAVs can use the on-board camera to navigate autonomously in unknown environments. Examples include interested target tracking, Available All degrees
navigation road following, and area coverage controlling. This project will develop software to identify interested objects from on-board camera
images and control the UAV accordingly. The developed algorithm can be tested in the robotic arena.

Mapping algorithm based Mapping 3D environment is a challenging problem for mobile robots due to the limitation of on-board sensors. Vision camera or laser Available All degrees
on vision camera or laser rangefinder can provide a scan of the environment. When it links to an aerial or ground mobile robot, it is possible to reconstruct a 2D or
rangefinder 3D map via merging multiple scans into a single map. But with the complex localisation problem, this mapping task is challenging. The
project is to investigate the techniques of solving this mapping problem. The reference can be found in the google open source
Cartographer (https://opensource.googleblog.com/2016/10/introducing-cartographer.html)
Hagras, Hani Interactive scheduling BT has developed a field force scheduling system which support automated, manual and hybrid task-resource allocation where field or Available All degrees
(industrial project with desk agent are guided by the system into the set of tasks they are advised to allocate next.
BT, co-supervised by Dr. Such system is interactive and to ensure a smooth adoption of the guidance generated by the system this is important to provide some
Anne Liret) explanation to the user (why does this action or this task is suggested as the best next one- how does it improve the overall schedule of
tasks or the overall productivity of the field agents or the service delivery, as example).
The project proposed to build a model that generates some explanations about the recommended next set of task allocations that the
scheduling system currently produces. This may be done using AI techniques or by simply simulating what alternatives actions would
produce and compare with the best option. This could as well be interleaved with a scheduling model is seen as suitable during the
project.

Skill requirements
Programming in java, some basics in operational research or scheduling is a plus.
Some knowledge in machine learning, CBR, fuzzy logic or other techniques of AI that may be used to generate explanation is a plus but
not required.

Business Process This project will examine the benefits and challenges of using deep learning to provide predictions on the evolution of graph structured Available All degrees
Predictions using Graph data. The specific use-case for these graph structures is the recording of temporal relationships in business process executions.
Structures in Deep Predictions can take the form of the next task that will occur, the time of the next task occurrence, the predicted final outcome of the
Learning (industrial business process, or some combination of these. Points of comparison with other predictive approaches can be provided.
project with BT, co-
supervised by Dr. Paul Skill requirements
Taylor) The student will need some knowledge of python, knowledge of tensorflow would be a plus but is not essential.

Refs:
- https://github.com/deepmind/graph_nets
- https://rd.springer.com/chapter/10.1007/978-3-319-71078-5_24

AI Explainability – AI Explainability – Argumentation Layer for resource scheduling: Available All degrees
Argumentation Layer for Our scheduling solution is responsible for allocating work to field technicians using a constraint-based model.
resource scheduling To do so it is optimising a multi-objective function using Local Search algorithms maximising allocation of tasks and minimising costs
(industrial project with (travel, quality of schedule, etc.).
BT, co-supervised by Dr. However to some extends this solution works as a black box that:
Raphael Dorne) - do not provide any explanation back to the end-user to justify any choice made,
- do not offer any mean to interact (argument) with the scheduler even though a lot of information is readily available inside or can be
easily derived from.

Hence the purpose of this project is to build an argumentation layer around this existing field force scheduling solution as depicted below.
This argumentation layer will be responsible for interacting with end-users in a human manner where proper conversations are expected.
Chatbots and Restful APIs technologies will be used to build this component.

Knowledge/Skills:
Programming skills in Java, Restful Web Services

Smart Object detection Automatic object recognition is a crucial activity for field service operations. For example, it could help novice users to identify specific Available All degrees
using AR (industrial pieces of equipment. When combined with Augmented Reality, could provide relevant information in real time to support engineers to
project with BT, co- complete their tasks. This project seeks to create an application for automatic object recognition using the Microsoft Hololens. This is a
supervised by Dr. Anasol collaborative project with BT, and as such, the content will be designed for field engineers tasks.
Pena-Rios) Deliverables
• Build an application for the Microsoft Hololens able to identify common objects in real time.
• Full documentation including technical documentation and user manuals.
Required technical skills and knowledge
• Programming skills in Phyton, Java and game programming (Unity, C#)

Smart Object detection Automatic object recognition is a crucial activity for field service operations. For example, it could help novice users to identify specific Available All degrees
on immersive 360 videos pieces of equipment. When combined with immersive 360 videos, it allows users to experience places or situations without the need to
(industrial project with physically move there. This project seeks to create an application for automatic object recognition on immersive 360 videos. This is a
BT, co-supervised by Dr. collaborative project with BT, and as such, the content will be designed for field engineers tasks.
Anasol Pena-Rios) Deliverables
• Build an application able to identify common objects in 360 videos in real time.
• Full documentation including technical documentation and user manuals.
Required technical skills and knowledge
• Programming skills in Phyton, Java and game programming

Big Data analysis using Opportunity to work on huge sensor data set from an IoT application using NoSQL & SQL databases. Available All degrees
NoSQL (Cassandra) & SQL · Develop a novel database structure to capture, store and access the sensor data.
databases for Internet of
things industry. In
conjunction with a
company called AGM.

Secure IoT Develop a private, secure and encrypted end-to-end network security using cryptography or SSL encryption for sending telemetry sensor Available All degrees
communication using data back to head office.
Virtual Private Network · Exploit developed IoT network by infusing man in the middle (MITM) attacks.
(VPN). In conjunction with
a company called AGM.

Visualisation of IoT data. Develop & design user friendly dashboard deployed on-premises to visualise telemetry sensor data. Available All degrees
In conjunction with a · Design custom and unique widgets for the dashboard.
company called AGM.

Application of ANN in Develop an artificial neural network to classify machine/pump faults like unbalance, looseness, bearing faults using vibration sensor. Available All degrees
Condition based
Predictive Maintenance.
In conjunction with a
company called AGM.

Forecasting water Model artificial neural network to forecast water consumption and the peak usage of water using pressure sensor & flow rate sensors Available All degrees
consumption using ANN.
In conjunction with a
company called AGM.
Halder, Sebastian Decoding mental imagery Brain-computer interfaces often employ imagery of motor tasks. Electroencephalography (EEG) signal is the most common option. Available Any
for a brain-computer However, it is beneficial to use additional measurement techniques that complement the EEG, like functional near-infrared spectroscopy
interface using infrared (fNIRS).
light and In this thesis you will use hardware available in the Essex BCI lab to record a dataset of combined EEG and fNIRS while participants are
electroencephalography performing mental imagery of motor tasks. Then you will use toolboxes written in Python to analyse the data. Finally, you will implement
This project will be co- an online filter based on this analysis. This work should ideally lead to a scientific publication.
supervised with Serafeim
Perdikis.

Using attention to tactile Brain-computer interfaces can be controlled using attention. Attending a Available Any
and auditory stimuli to particular stimulus causes changes in brain activity in response to that
write with a stimulus that can be recorded from the surface of the scalp. These
brain-computer interface signals are recorded using electroencephalography (EEG). The stimuli the
user attends to can be auditory, visual or tactile. In this thesis you
will use the measurement system available in the Essex BCI lab to record
a data set of combined auditory and tactile stimuli. Then you will use
toolboxes written in Python to analyse the data which you will then
classify into attended and non-attended stimuli using machine learning.
Finally, you will implement an online filter based on this analysis.

Controlling a virtual race The University of Essex will participate in a race, called the Available Any
car with your thoughts Cybathlon, in which a paralysed pilot will control a virtual race car
This project will be co- using only brain signals. These brain signals are recorded from the
supervised with Ana scalp and the pilot controls them by performing particular mental tasks.
Matran-Fernandez In this project you will record data while participants are performing
the same tasks as the pilot in the Cybathlon. Then you will use machine
learning and signal processing tools written in python to analyse the
data and produce a signal processing pipeline that can be used to
control the actual virtual race used in the Cybathlon race.

Measuring the complexity Using short, magnetic pulses the brain can be safely stimulated to Available Any
of brain responses using produce simple responses such as a muscle twitch. The way this response
magnetic stimuli occurs over time and space can be used to, among other things, determine
This project will be co- the state the person being stimulated is in. For example to determine if
supervised with Ian Daly someone is conscious or unconscious. In this project you will use a
purpose built robot to magnetically stimulate the brain and record the
responses using electroencephalography. Second, you will transform these
responses into a virtual representation of the brain using Python and
then analyse the responses to calculate a measure called the
perturbational complexity index, which measures the "complexity" of the
brain activity and may be used to measure levels of consciousness.

Investigation of the Chronic pain affects a large number of people in our society. A better Available Any
effects of hot water understanding of how pain is represented in the brain may allow us
immersion on brain improve how pain can be treated. In this project you will use
signals electroencephalography and functional near-infrared spectroscopy to
This project will be co- record data from the brain while the participant has one hand immersed
supervised with Elia in hot water. You will then use machine learning and signal processing
Valentini (Psychology) algorithms in Python to analyse the data to determine if the brain
signals were affected by the hot water immersion.

Learning to spell with a Spelling with a brain-computer interface can be a challenging task. In Available Any
brain-computer interface this project you will investigate new methods of machine learning on
existing data of people writing words with a brain-computer interface
based on visual and auditory stimulation. Your goal will be to reduce
the number of stimuli required before the correct letter is recognised
by the system to the minimum amount possible. Ideally your algorithm
will provide reliable feedback to the user after every stimulus
presentation. To achieve this goal you will use machine learning and
signal processing algorithms in Python. After this offline analysis you
will implement your method in an online filter designed for a
brain-computer interface system. You will then evaluate how well your
filter can be used for spelling with this brain-computer interface.

Detection of After a brain injury or during surgery it may not be possible to Available Any
consciousness using the communicate with a person but it is still critical to know whether they
spontaneous are conscious or not. You will signal processing techniques that can be
electroencephalogram applied to the spontaneous electroencephalogram (EEG), that means
without external stimulation, to determine if a person is conscious. The
methods suitable for this include measures of complexity, connectivity
and the power spectral density. You will implement an algorithm using
these techniques and machine learning in Python. After the approach has
been sufficiently refined you will modify the algorithm to be applicable
in a real-time recording.

Hoshiar, Ali Small scale robots for Significant progress in micro/nanoscale science and technology in last two decades has created increasing demand and hope for new Available Any
health care microsystems for high-impact applications in healthcare and biotechnology. Such microsystems should be able to access small-enclosed
spaces such as inside the human body and microfluidic devices noninvasively and manipulate or interact with micro/nanoscale entities
directly. Because human or macroscale robot sensing, precision, and size are not capable of achieving such desired characteristics,
microrobotics has emerged as a new robotics field to extend our interaction and exploration capabilities to sub-millimeter scales. Despite
varies applications, their nano intuitive physics, limited actuation, sensing, computation, and communication, difficult observation, and
time consuming expensive fabrication are only a few limiting bottlenecks. Molding, fabrication, design and implementation of these
systems can be studied.

Soft robotics Most living organisms have soft interactions with the environment, and soft robotics is a new subfield of robotics mimicking this behavior. Available Any
In this field the design, fabrication, modeling, characterization, and control of soft robots are studied. The soft robots have divorce
applications from manipulation of fragile objects (like soft fruits) to medical applications in angioplasty and colonoscopy. Fluidic,
pneumatic, and magnetic are most popular actuation methods for soft robots. The goal of this research is to develop miniature soft
robots for medical applications as well as soft manipulators for object manipulation. More information and relevant publications are
available upon request.

Swarm robotics at small- Swarm robotics study collective behavior of robots in-group mimicking swarm behavior in nature. Small-scale robots for medical Available Any
scale treatments like targeted cell/drug delivery, and hypothermia recently emerged. This approach aims to generate an optimal
concentration of curing agent; i.e., a level greater than the minimum required and lower than the harmful dose, at the target location in
the body. For the treatment to be successful the swarm control and implementation of these agents is at most important. The main
research objective is to design, fabricate and implement swarm of robots for small-scale applications. More information and relevant
publications are available upon request.
Data driven design and In many recent robotics applications like micro/nano robotics and soft robotics, mathematical modeling of the plant is considered hard to Available Any
control achieve. With the rapid developments in data science, the data-driven methods are proposed to solve this issue. In this approach a model
is fitted to the experimental/simulation data collected. This model later is exploit to design and control the system. Although, it is still
difficult to find a simple yet reliable model for a physical system to fully cover the system’s complex behavior, it significantly improves the
system performance in many applications. This approach can be used for already available experimental setup. The approach also can be
explored for both modeling and implementation on soft robotics and small-scale robots. More information and relevant publications are
available upon request.

Mechatronic systems for Mechatronics is a combination of mechanical, electrical and computer engineering used for developing intelligent systems. Small-scale Available Any
small-scale robotics robotics mostly rely on magnetic field generated by permanent or electromagnets. Mechatronics systems are used to generate and
control these magnetic fields. These newly develop systems are of great interest to robotics community and many challenges like higher
controllability, region of interest, non-homogeneity in the field, dynamic magnetic actuation for field functionalization are only few topics
to study. The goal of this research is to develop mechatronic systems for small-scale robotics. More information and relevant publications
are available upon request.

3D and 4D printing for The additive manufacturing (3D printing) is the process of fabricating three-dimensional objects from a computer-aided design (CAD) Available Any
mechatronic applications model by adding material layer by layer. Combining this fabrication techniques and mechatronics engineering, enabled us to fabricate
almost any device at low cost. More recently, 4D printing concept is introduced. In 4D printing different materials are used in the
structures and actuated in response to different environmental conditions. The applications of 4D printing in health care and
environmental studies are subject of this research. The alternative objective of this study can be use of 3D printing techniques combined
with mechatronic engineering to fabricate low cost systems ex. haptic device, syringe pumps, etc. for applications in medicine and agri-
tech. More information and relevant publications are available upon request.

Hu, Huosheng; Gesture based control of This project will be focused on Draganflyer SAVS, i.e. a gyro stabilized RC helicopter with an on-board stabilized aerial video system. It has Available All degrees
an unmanned aerial four rotors, also namely X4-flyer. The task of this project is to investigate how an UAV could be controlled by human hand gestures. A 3D
vehicle Kinect sensor will be used to capture human arm motions and recognize human intensions for the control purpose. The human gestures
will be used to control the UAV motion and implement a specific task, e.g. tracking a radio-controlled toy car.

Smart wearables system The last decade has witnessed a rapid surge of interest in new sensing and monitoring devices for healthcare and the use of wireless Available All degrees
with wireless distributed wearable devices. Applications and practical deployment of such devices/sensors include elderly care, managing patients with chronic
sensors. diseases, rehabilitation, sports, and entertainment. The student will use several healthcare sensors such as ECG, PPG, temperature, and
MEMS. Further sensors will be explored. These sensors will be interfaced via bluetooth 5 and will communicate to central bluetooth
device. Afterwards, data will be collected to produce a health report.
This is an industry-based project, and Aakash Soni will be Company Supervisor at August International Ltd., which is located at Hoddesdon,
Hertfordshire.

Cloud computing for Cloud computing provides abundant resources and computing power through wired or wireless networks. This project aims to create a Available All degrees
monitoring fall detection cloud computing system for monitoring a large number of wristbands for fall detection of the elderly people at individual homes. The fall
of the elderly at homes detection results are sent to the cloud computing system via smartphones for emergency responses and cloud services.

This is an industry-based project, and Dr Hongtao Zheng will be Company Supervisor at August International Ltd., which is located at
Hoddesdon, Hertfordshire.

Machine learning for This project aims to deploy machine learning techniques to improve the performance of a fall detection algorithm that runs on a Available All degrees
reducing false alarm of smartwatch. Data collected from fall events will be used to train the algorithm so that the false alarm rate could be reduced.
fall detection Project Breakdown:
• Implement 9-axis sensors with nRF52840 board.
• Use quaternion approach
• Implement Madgwick’s filter
• Develop machine learing algorithm for fall detection
This is an industry-based project, and Dr Hongtao Zheng will be Company Supervisor at August International Ltd., which is located at
Hoddesdon, Hertfordshire.

Human Adaptive This project aims to develop visual based human-robot interface to control a Baxter robot. It will enable the transfer of human Available All degrees
Manipulation of Baxter manipulation intentions into a two-arm robot to allow its manipulation within a domestic environment. The project’s objectives are:
Robot 1) To deploy 3D Kinect sensor for collecting intention data from human upper limb gestures, and control the robot arms to mimic human
arm movements.
2) To deploy classification algorithms to identify user manipulation via the anatomy of the human arm.
3) To develop on-line learning methods for tuning the controller so as to adapt to specific users and changing environments for the
efficient manipulation.
4) To validate the proposed algorithms through a set of lab experiments, and evaluate potential impacts.

Hu, Liang Evolutionary algorithms This is practical project that needs algorithm development and simulation research. In this project, the student will be developing Available All degrees
based online path autonomous real-time path planner for autonomous surface vehicle using evolutionary algorithms like GA and PSO. The path planner
planning for autonomous should be able to plan a local path to avoid collision with other vehicles while keep in compliance with traffic rules at sea, namely
marine vehicles COLREGs. The performance of different evolutionary algorithms based path planner should be demonstrated and compared in simulation.
The student should be familiar with computational intelligence or heuristic-optimisation and is also confident in using Matlab, Python or
other languages. Reference: Liang Hu et al. "COLREGs-Compliant Path Planning for Autonomous Surface Vehicles: A Multiobjective
Optimization Approach." IFAC-PapersOnLine 50 (1), 13662-13667, 2017. If you have any question about this project contact
l.hu@essex.ac.uk

Colregs-compliant This is a practical project that needs algorithm development and simulation research. In this project, the student will be developing and Available All degrees
Collision avoidance for implementing autonomous real-time path planner algorithms for autonomous surface vehicle using reciprocal velocity obstacle (RVO)
autonomous surface approach. RVO is a fast path planning algorithm widely used for reactive navigation and there are also many variant algorithms of RVO
vehicle using reciprocal (Many open-source codes can be found online). You can choose one variant of RVO and compare it with the basic RVO in this project. The
velocity obstacle path planner should be able to plan a local path to avoid collision with other vehicles while keep in compliance with traffic rules at sea,
approaches namely COLREGs. The student should be familiar with computational intelligence or heuristic-optimisation and is also confident in using
Matlab, Python (preferred) or other languages. Reference: Paravisi, Marcelo, et al. "Unmanned Surface Vehicle Simulator with Realistic
Environmental Disturbances." Sensors 19.5 (2019): 1068. If you have any question about this project contact l.hu@essex.ac.uk.

Experimental assessment This is a development project that needs software tests and assessment. The use of robotics in disaster scenarios has become a reality Available All degrees
and development and it is important to have a high-fidelity simulator with realistic USV models where designers can assess their control strategies under
autonomous surface different degrees of environmental disturbances. Recently quite a few open-source USV simulators have developed for USVs (see the
vehicle simulation references for more details), integrated with robotic middleware like Robot Operating System (ROS).This project is 1) to develop various
platforms simulation scenarios such as maps, obstacles and hydrological model; 2) test and evaluate ASV’s control, mapping, and task-specific
algorithms in different simulation platform; 3) provide critical comparative assessment of different simulation platform and a
recommended benchmark simulation scenarios. The student should has some basic knowledge of mobile robots and AI and is also
confident in using Matlab or Python and robotic simulators ROS or VREP. Reference: Kuwata, Yoshiaki, et al. "Safe maritime autonomous
navigation with COLREGS, using velocity obstacles." IEEE Journal of Oceanic Engineering 39.1 (2013): 110-119. If you have any question
about this project contact l.hu@essex.ac.uk
Deep reinforcement Event-triggered control (ETC) methods can achieve high-performance control with a significantly lower number of samples compared to Available All degrees
learning for event-based time-triggered methods. Recently deep reinforcement learning (DRL) algorithms have been proposed to simultaneously learn control and
control of robotics using communication behaviour that is particularly suitable for ETC. This projects aim to extend the existing DRL algorithm for adapting to
OpenAI Gym other different event-based control strategies, and compare it to model-based event-triggering frameworks. In particular, the algorithms
are to be validated on robot locomotion task in OpenAI Gym using physics simulation engine, which can be generated as the project
demos. The student should has some background of robots and control and research interest in deep reinforcement learning and is also
confident in using Python. Reference: Baumann, Dominik, et al. "Deep reinforcement learning for event-triggered control." IEEE
Conference on Decision and Control (CDC). 2018. If you have any question about this project contact l.hu@essex.ac.uk

Safe control for Safe autonomy is important in many application domains, especially for applications involving interactions with humans. We should ensure Available All degrees
interactive robot design not only the safety of the ego robot, but also the safety of other agents (humans or robots) that directly interact with the autonomy.
Recently, many energy function based safe control algorithms have been proposed and tested through benchmark simulations. This
project aims to develop and implement safe control algorithms in the mobile robots (there are many choices in Robot Athena), and test
and demonstrate it in a real-world indoor environment with moving obstacles. The student should have background in control
engineering and/or robotics control and is also confident in using Matlab, Python (preferred) or other languages. Reference: T. Wei, and
C. Liu “Safe control algorithms using energy functions: A unified framework, benchmark, and new directions.” IEEE Conference on
Decision and Control (CDC), 2019. If you have any question about this project contact l.hu@essex.ac.uk

Welcome students to self- Please make an appointment to discuss possibilities via l.hu@essex.ac.uk Available Prerequisites and
suggestion on a project in requirements: MSc in
the area of learning for Artificial Intelligence,
robotics and control, MSc in Robotics and
either algorithmic or Intelligent Systems, MSc
application research. in Data Science. Excellent
programming skills
(Matlab, Python), interest
and knowledge on
applications of machine
learning and AI in real-
world problems.

Jarchi, Delaram Deep learning for speech Vairous Automatic Speech Recognition (ASR) techniques are proposed aiming to improve the noise robustness and speech enhancement. Available Any degree
recognition This project aims to explore very recent speech enhancement techniques using deep learning techniques such as long short-term memory
(LSTM) and convolutional neural network (CNN) exploting advanced time-frequency analysis techniques.

Sleep analysis using deep Classification of sleep stages is crucial for diagnosis and treasing sleep disorders. Recent methods exploit deep learning approaches for Available Any degree
learning approaches automatic sleep stage classification. In this project, the student will explore public datasets to implement deep learning approaches as
such as convolutional neural network (CNN) using data recorded from electroencephalography (EEG), electromyography (EMG) and
electrooculography (EOG) to compare the performance of deployed deep learning techniques.

A generative model for Training predictive models to diagnose heart disease using electrocardiogram (ECG) signals requires a large volume of labeled clinical Available Any degree
ECG signals using deep data. An interesting technique has been recently published in Scientific report journal, where a deep learning technique is producing ECG
learning signals with similarity to real ECG signals. The proposed method uses generative adversarial network (GAN) model, and is composed of a
bidirectional long short-term memory (LSTM) and convolutional neural network (CNN). Implemetaion of a hybrid deep learning technique
is required in this project where the MIT-BIH arrhythmia dataset can be exploited.

A deep learning approach Respiratory rate (RR) which is simply the number of breath per minute and is an important vital sign that has been used inside intensive Available Any degree
for extraction of care units to detect deteriorating patients and avoid death. Estimation of RR is also useful for monitoring patients with breathing
respiratory rate from disorder e.g. patients with asthma or chronic obstructive pulmonary disease (COPD). Automatic and continuous estimation of RR from
optical sensor optical sensors is significantly useful to complement clinical decisions. The aim of this project is to apply deep learning techniques for
robust estimation of RR using optical sensors.

Heart rate analysis from Heart rate estimation during motion is a challenging task since the signals from wearable sensor contain motion artefacts. Artefact Available Any degree
optical sensor during removal from optical sensor is necessary as a pre-processing step for better estimation of heart rate. In this project, the student will use
motion various signal processing techniques to remove the motion artefact of the optical sensor with the aid of a simultaneous motion sensor
(using accelerometer). The IEEE signal processing cup dataset 2015 which contains simltaneous electrocardiography (ECG),
photoplethysmpgraphy (PPG) and accelerometer data including reference heart rate as ground truth.

Deep neural network to Blood pressure as an important vital sign is direct indicator of hypertension that can contribute to heart attacks and strockes. Cufless Available Any degree
estimate blood pressure estimation of blood pressure from wearable sensors has attracted many research studies recently. In this project, the student will explore
from optical sensor deep learning techniques to estimate blood pressure. MIMIC dataset which is available on phyionet website will be used for validation of
machine learning based paltforms for estimation of blood pressure using optical sensor.

Karapetyan, Daniel Automated design of Optimisation algorithms are at the core of many decision support/making systems. In many applications, only bespoke algorithms can be Available All degrees.
optimisation algorithms used. Development of bespoke optimisation algorithms is resource expensive and time consuming, and hence there are a lot of benefits Requirements: strong
of automated generation of optimisation algorithms. This project will develop ideas of a recent approach to automated generation of programming skills
optimisation algorithms based on Markov chains. (C/C++, Java or
C#), knowledge of
standard data structures
and algorithms

Data-driven analysis of While running, optimisation algorithms generate large quantities of data. This data reflects properties of the problem instance and the Available All degrees.
optimisation algorithm algorithm itself. Hence, this data can be used to analyse the algorithm's performance and drive decisions within the algorithm. This Requirements: strong
performance project will investigate data science methods that can efficiently extract useful information about algorithm's performance and verify programming skills
how such an information can be used. (C/C++, Java or
C#), knowledge of
standard data structures
and algorithms

User preference learning Routing (searching for a best route from A to B) is a relatively simple mathematical problem which is successfully solved by many Available All degrees.
in routing applications services and software such as Google Maps or Garmin. However mathematically optimal (shortest/fastest) routes are not always ideal Requirements: strong
from the user's point of view. This project will investigate how software can learn user preferences in choosing a best route and what programming skills
relevant information can be extracted from open access maps. (C/C++, Java or
C#), knowledge of
standard data structures
and algorithms

A chatbot for solving Chatbot is a convenient interface between a user and a computer system. While some chatbots are designed to provide access to data or Available All degrees.
problems control devices, this project will develop a chatbot that makes use of the mathematical power of computers. As a simple example, it may Requirements: strong
work like a calculator or unit converter. It may also be able to do do ordering of tasks in an optimal way, timetabling, etc. The system will programming skills
be extendable to allow adding new functionality in future. (C/C++, Java or
C#), knowledge of
standard data structures
and algorithms
Interpretable player in 2048 is a simple yet addictive videogame (http://2048game.com/). It requires a long term strategy yet playing it is relatively easy. In this Available All degrees.
the 2048 videogame project, you will investigate how an efficient playing strategy can be described in an interpretable way. This strategy will need to be Requirements: strong
learnt by your software. programming skills
(C/C++, Java or
C#), knowledge of
standard data structures
and algorithms

User-friendly tool for semi- Due to the many approaches to clustering, it is often difficult for a non-specialist end user to apply the right tool for their job. This project Available All degrees.
supervised clustering aims at developing a user-friendly service that would engage the user into an easy to follow interactive process. During this process, the Requirements: strong
tool will learn a distance metric for the provided data and then use it to cluster the data and visualise the results. programming skills
(C/C++, Java or
C#), knowledge of
standard data structures
and algorithms

Kolozali, Sefki Making sense of sounds: This project aims the detection of scenes and events from environmental sounds. Sound is often a useful complement to modalities such Available Any degree
detection and classifi- as video, carrying information not otherwise present such as information from speech and birdsong. It is easy and more convenient to
cation of acoustic events collect by using different devices such as mobile phones. Having semantic audio information can be useful for further processing such as
robot navigation, user alerts, or analysing and predicting patterns of events. Moreover, the same technologies have also applications in
cataloguing/searching audio archives, whose digital collections have grown enormously in recent decades.
The candidate will develop an acoustic classification system to detect the events from a given environmental sound dataset. The
candidate must be familiar with pattern recognition techniques and python programming language. The candidate will be using a publicly
available environmental sound data set.

Facial emotion This project aims developing a facial emotion detection system by using pattern recognition techniques. The detection of human Available Any degree
recognition from videos emotions plays a crucial role in un- derstanding human behaviour and health conditions. With the recent advancements in sensor
technologies and machine learning, it is possible to undertake remote mon- itoring at scale now. Having such emotion recognition
systems can help us better understand the behavioural patterns and their role in different health problems, such as obesity and mental
health. Early detection of such health conditions could help to prevent extreme measures by notifying people in advance by using
technological devices.
The candidate will develop a probabilistic method to extract information from detected face and distinguish the emotions in a given
video. The candidate will be provided with a publicly available emotion data set. The candidate must be familiar with basic pattern
recognition techniques and Python programming language.

Detection of Physical Recent technological advances made it possible to undertake remote monitoring at scale. The detection of physical activities plays a Available Any degree
Activities using Wearable crucial role in understanding human behaviour and health conditions. It can help us better understand the behavioural patterns and their
Devices role in different health problems, such as obesity and mental health. Early detection of such health conditions could help to prevent
extreme measures by notifying people in advance by using technological devices
However, since some of the activities may involve similar movements, such as brushing teeth and eating, it can be very challenging to
distinguish these activities. This project is about the detection of physical activities of people (e.g. eating, walking, running,
resting/standing, going up and down stairs, sitting, sleeping) using multiple sensors attached to a wearable device, and make sense of the
phenomena in the micro-environments. The candidate must be familiar with pattern recognition techniques and Python programming
language. The candidate will be provided with a cutting-edge wearable device to collect his/her data set.

Change detection and The project aims the development of data analysis techniques to detect change or fault in a given multivariate time series data. With the Available Any degree
notification system increasing number of available continuous time-series data sets in different domains (e.g. health, finance, internet of things), it is
becoming more crucial to detect anomalies and change in (near) real-time. The change detection systems are regularly used in fault
detection of factories, in the communication strategy of smart devices, as well as health and finance applications. The candidate will
develop a change detection system using probabilistic/statistical methods and will be provided with multivariate time series data sets.
The candidate must be familiar with basic pattern recognition techniques and Python programming language.

Li, Junhua Understanding of brain In this project, the student is expected to explore neurophysiological signals (i.e., EEG) recorded from human brain, develop programmes Available Any degree
connectivity pertaining to using a programming language you are familiar to estimate connective strengths, distinguish the differences exist among different
emotion emotion categories (e.g., happiness and sadness), and classify emotion categories based on the features representing the differences
exist among emotion categories. The objective of the project is twofold: (1) reveal differential differences among emotion categories
from the perspective of brain connectivity; (2) classify emotion categories based on the neurophysiological signals.

Visualisation of intrinsic The project aims to visualise different kinds of characteristics containing in EEG signal relevant to emotion. The characteristics should be Available Any degree
characteristics of emotion first explored from temporal, spectral, and spatial domains, as well as their combined domains. Then, the form of visualisation is designed
to show these characteristics in an attractive way. It would be expected to visualise them on a mobile portal device such as a cell phone.

Emotion recognition using Emotion recognition plays an essential role in the development of intelligent robots and the creation of a smart environment. Machine Available Any degree
sophisticated methods learning methods could achieve emotion recognition. In this project, the modified model aiming at recognising emotion is required to be
developed based on a comprehensive understanding of the-state-of-art methods which have been utilised in emotion recognition. The
objective is to propose a modified model, which has a better performance in emotion recognition.

Enhancing cognitive When people become old, cognition would be declined more or less. This project is to find out whether cognitive capacity could be Available Any degree
capacity through the preserved by a certain amount of training by cognitive tasks. The data collected from older adults will be provided. You are expected to
training by cognitive tasks analyse the data to discover the effect of the cognitive training and answer the above question according to the results derived from the
analyses.

Brain connectivity-based You will engage in EEG processing, feature extraction, and classification using a variety of methods. You will also need to employ deep Available Any degree
classification of EEG for learning model for the EEG classification. It is recommended to make some modifications of the existing deep learning models to improve
brain-computer interface the classification performance in the EEG classification.

Workload classification Workload can be measured through the signal of EEG. This is able to help people organise their work efficiently and make a better Available Any degree
based on deep learning balance between work and rest. You are expected to develop a deep learning model to classify different levels of workload. You will use
models your familiar programming language to code the deep learning model and implement the steps from EEG processing to final classification
and report the performance results of comparisons between the deep learning methods and other methods.

Matran-Fernandez, Development of an See: https://ieeexplore.ieee.org/abstract/document/6824163/ Available Any degree


Ana automated spike sorting https://link.springer.com/chapter/10.1007/978-3-540-74494-8_77
algorithm for neural https://ieeexplore.ieee.org/abstract/document/4291854/
recordings (project co- https://www.hindawi.com/journals/np/2016/3489540/abs/
supervised by Prof Luca
Citi)
Creation of a method to One of the common issues in the analysis of electroencephalographic (EEG) signals is the presence of noise. In this project, you will create Available Any degree
clean EEG signals based an autoencoder and train it using databases of EEG signals freely available. The autoencoder should learn to clean the signal by filtering
on autoencoders out the noise and reconstructing the information present in the EEG signal.

A Brain-Computer The University of Essex will participate in a race, called the Cybathlon (https://cybathlon.ethz.ch), in which a paralysed pilot will control a Available Any degree
interface system for virtual race car using only brain signals. These brain signals are recorded from the scalp and the pilot controls them by performing
Cybathlon particular mental tasks. In this project you will record data while participants are performing the same tasks as the pilot in the Cybathlon.
Then you will use machine learning and signal processing tools written in Python to analyse the data and produce a signal processing
pipeline that can be used to control the virtual race car used in the Cybathlon. This project will be co-supervised with members of the BCI
lab.

Student-proposed project Pleaase arrange a meeting to discuss Available Any degree


in Brain-Computer
Interfacing/Neural
Engineering

Detecting affect from You will design an experiment and use it to collect electroencephalographic (EEG) data from participants while they watch streams of Available Any degree
electroencephalography images with affective content. You will then analyse the data to determine whether the type of content (positive or negative) of the
in rapidly presented streams can be decoded from the EEG.
streams of images

Mirshekar, Dariush Steerable antennas for 60 In this project, a planar steerable antennas using phase shifters based on liquid crystals is to be designed. If time permits, it will be Available Telecommunications
GHz radio systems fabricated and measured. The antenna element in this antenna array can be a microstrip patch or a slot. The project involves liquid
crystal physics, advanced electromagnetic theory, microwave and antenna theory and technology.

Characterisation of Periodic structures can provide filtering function at microwave frequencies. In this project, the objective is to characterise various two- Telecommunications
microstrip periodic dimensional periodic structures using different microstrip resonators for 60 GHZ applications. The project involves advanced
structure for 60 GHz electromagnetics, microwave theory & applications, high frequency circuit theory & design, and fabrication.
applications

UWB filters A work started recently is to be pursued in order to synthesize miniaturised filters for UWB and other applications. In this work use is Telecommunications
made of slotted ring resonators or microstrip E-shape structure. The project is mainly theoretical involving advanced electromagnetics
and circuit theory & applications.

Adaptive filters using MEMS (Micro-electro-mechanical switches) has been used in optical communications for channel selection etc. In this project, a feasibility Telecommunications
MEMS study will be carried out to establish potentials of MEMS in design of mm-wave filters in band adaptive applications. The project involves
advanced electromagnetics, RF engineering, high frequency circuit theory and computer simulation.

High gain multilayer For satellite communications, high gain antennas of high polarisation purity is normally required. To this end, in this project, the potential Telecommunications
microstrip patch of multilayer patch antenna array will be investigated, and if time permits, a suitable design will be fabricated and tested. The project
antennas involves advanced electromagnetics, RF engineering, antenna theory and technology, and computer simulation.

Mohan, Fruitbot- Collaborative This project will build up on our existing research on Agricultural robotics (mainly soft fruit detection, picking, and packing) in collaboration Available Any degree
Vishwanathan Robot for Soft Fruit with one of the leading growers in the country Tiptree farms. The project will developed on the state of the art bimanual mobile
Harvesting collaborative robot ‘The BEAST’ designed and assembled at Essex though a £80K strategic equipment grant. The BEAST is a unique
platform that incorporates Husky UGV carrying two UR3 arms, a modular soft gripper, a specially designed Gripper/Cutter, 3 Stereo
cameras, Two deep learning cameras and Velodyne Puck Lidar. The robot is presently located in the new lab adjacent Lab 7. The project
lies at the intersection of cutting edge Perception, Robotic Manipulation, Machine learning, Navigation and some level of human robot
collaboration.
The project is available for two highly motivated students- one focusing on the perception side and other on the action side.
The AgriFood sector is under severe pressure due to several interlinked reasons like ageing population, critical shortage of labor to
perform laborious, repetitive tasks like picking/packing etc, Increasing demand, need to reduce production cost and environmental
sustainability. Our goal is to provide a technological solution to this problem. The challenge is that in comparison to a typical industrial
work cell, the agricultural environment is completely unstructured-hence requiring adaptivity, precision and intelligence in both
perception and action.
On the perception side- the challenge is to identify/localize partially occluded berries in the plant while also taking into account changes
in the environment (lighting conditions etc. typical of a farm) using machine learning (we will use the ZED camera). On the action side,
the project will further develop our cutting edge neural motion controller (with an accuracy of 3 mm presently) for dexterous bimanual
coordination, soft manipulation of strawberries. The perception side will also explore clever machine vision/learning techniques to
predict/estimate the weight of the berry through vision (Point cloud information) to support insitu packaging. Packing onsite has the
potential to eliminate the need for separate pack houses in the farm leading to significant cost reductions.

Smart Robots for Smart This project will focus the deployment of a state of the art humanoid (Pepper) in the newly opened smart home (iSpace), to develop a Available Any degree
Homes cogntive robotic companion for elderly care, people with mild cognitve impaiment (like dementia). The rapidly increasing percentage of
ageing population and large number of individuals with long-term conditions like such as stroke, dementia (among others), the prohibitive
costs of 24/7 monitoring, shortage of caregivers is presently putting severe pressure on the health care system, and family of affected
individuals. That advances in both hardware, software and cognitive capabilities of robots can serve as force multipliers for elerly
assitance and social care is now recognized by several robotics roadmaps (EU H2020). Smart Robots in Smart homes is an adventrous
exploration in this direction.
The Robotic companion (Pepper) will be able to control the smart home (iSpace) through voice, provide personalized assitance to users
(remember special needs of users, fetch objects, monitor the environment and alert caregivers in case of emergency, provide
compnionship). The project is very broad (and open to students own interest and creativity). Some of the areas to focus are-
Enhanced Interaction and Social Intelligence- enabling Pepper to interact with the smart home and the user though a combination of
speech, tablet, movements and gestures (and exhibit apprprite behaviours based on user requests).
Enhanced Mobility- development of a landmark based navigation architecture with swift learning of new environemnts/homes (and
adaptation when things change)
User Modelling for Personalized assitance- Learn about the user preferences and encode this information into the robot episodic memory,
that will enable recalling information based on partial cues (for example- recognizing the user by face and remembering the time for
his/her medication, where the medicine is and fetch it).

Self Suggested proejcts to You can suggest a project of your own or discuss with me! Beast is a new custom designed robotic platform assembled at Essex through a Available Any degree
enhance ‘Perception- £80K strategic equipment grant. The robot is presently located in the new lab adjacent Lab 7 which forms a part of the ECPIC (Essex Plant
Action- Inteligence’ of the Innovation Centre).The BEAST incorporates Husky UGV carrying two UR3 arms, a modular soft gripper, a specially designed
new Beast Robot Gripper/Cutter, 3 Stereo cameras, Two deep learning cameras and Velodyne Puck Lidar. The project lies at the intersection of cutting
edge Perception, Robotic Manipulation, Machine learning, Navigation and some level of human robot collaboration.
The Beast is envisaged to be a platform that will enable several cutting edge applied and research projects related to autonomous
navigation, computer vision on a mobile autonomous robot, bimanual coordination, fine manipulation including assembly/tool use in
robots, reasoning and planning, human robot collaboration. These are core features needed for future robotic ‘coworkers’ assisting us in
several domains- domestic, industry, extreme environments. See http://www.robocupathome.org/ , http://www.robocupatwork.org/ .
So I am happy to receive your ideas on what you want to do with the Beast (and we can converge on a suitable theme that is impactful
and novel)! In particular, I am interested in
Mobile Bimanual coordiantion-Coordinating body movements and the two arms for goal directed manipulation/assembly.
Autonomous Navigation- combining information form Velodyne Puck Lidar, Stereo cameras and neural networks for human aware motion
planning.
Active Perception- Using one hand to declutter the environemnt and the other to manipulate. In this case the predictions/actions of the
robot changes its perception (what it sees or could see).
Human Robot Colllaboration- The robot has three stereo cameras and two machine vision cameras . The idea is to use them to detect
human actions, predcit intentions/potential actions and drive collaborative behaviours.
“Being in the Others During any social interaction, cooperation (with someone else) we are both acting and at the same time predicting the consequences of Available Any degree
Shoes “-Perspective our actions, others actions/goals. This ability to take the perspective of the other (infer his/her potential goal) is a critical desirable
taking for Joint Goal feature for effective human-robot, robot-robot collaboration, thus opening up applications in several domains: industry, domestic among
‘human-robot’ others. In fact, the present trend is that the field of Robotics is gradually moving towards Cobotics!
collaboration In this context, this project will explore this critical open problem in social robotics and develop a simple perspective taking system,
implemented on the UR3 robot available in the Robot Arena. Perspective taking will be explored/implemented at two levels 1) Short
term perspective taking: recognizing the action of the collaborating human and inferring the possible goal; 2) Long term perspective
taking: combining information from (1) with the robots own memory/experience to make long term predictions (like possible next action
sequences). Simple but engaging scenarios like human picks an empty cup-pour water into it or human throws a piece of bread in the bin
(goal-S/he is cleaning the table) will be used to demonstrate the project. The system has direct applications in all areas where robots
have to work alongside humans- from manufacturing to agriculture.
Good supporting software, background literature developed from previous European Union funded project DARWIN (Darwin here stands
for Dexterous Assembler Robot Working with embodied INtelligence) and EFAA is available to get started.

The Storm Cone AR art (in The Storm Cone is an ambitious new site-specific spatial sound installation by award winning artist Laura Daly. It will trace lost Available Seeking a skilled and
conjuction with Laura bandstands in England, using geo-locative & immersive technologies. Experienced via mobile devices & headphones, audience movement dedicated Developer
Daly, artist) will trigger & alter the 3D sounds of newly commissioned brass ensemble music by award winning composer and conductor Tom Davoren (Unity) to play a vital role
& eight adjacent digital sound works by Laura Daly, in & around the sites of lost bandstands. in the production of an
A 3D AR line drawing of a bandstand will be positioned on the site of a lost bandstand, using GPS. Visible via the viewers hand held device exciting new AR art
as they approach the site, they will experience the interior and exterior of the structure. Entering the AR bandstand, the sound of the project using 360 sound
commissioned brass music will be triggered. Perception of the bands individual instruments will change as the device is moved around the and AR vision – The
‘bandstands’ interior. As instruments progressively fade away during the commissioned music, small AR line drawings of each instrument Storm Cone. Joining a
will appear in the vicinity of the bandstand - leading the individual to the eight different sound works placed around the park. As they talented team of artists,
reach each AR icon, they will hear the individual instrument ebb into a sound work (triggered by GPS). A fuller project outline is available musicians and
and there may be an opportunity for some of this work to be paid. collaborators, the initial
launch will be late
summer/early autumn
2020 in NetPark,
Chalkwell Park Southend
(Estuary Festival) and
Peel Park, Salford
(Rediscovering Salford).

Musavian, Leila Interference Efficiency in One of the major challenges of the future wireless communication networks is the scarcity of available spectrum. There are several Available IoT, Telecommunication
Cognitive radio channels proposals on how we can tackle this problem, and cognitive radio technology is one of them. The principle of cognitive radios is that the and data
users with no licence to the channel (referred to by secondary users) can access the spectrum as long as their operation does not communications
harmfully affect the licenced users (referred to by primary users) of the channels. However, there are several challenges to making this programmes
technology a possibility of the future.

In this project, we investigate a novel performance metric, called interference efficiency, which shows the number of bits versus the
imposed interference energy. This metric is particularly in line with the design of secondary users networks wherein we aim to improve
the rate of the secondary users while reducing the interference level at the primary receivers. We aim to analyse the performance of a
single secondary user considering various design parameters.

Strong mathematical background and good understanding of the modulation techniques, system performance and probability and
statistics are required.

Performance analysis of This project is to analyse the performance of non-orthogonal multiple access (NOMA) techniques, which is a part of the suggested Available IoT, Telecommunication
NOMA platform for the 5G systems. Implementing NOMA still faces several challenges, and one of them is that the performance of NOMA in and data
different realistic environment scenarios is still unknown. Different modulation schemes are implemented in the current communication communications
systems. The aim of this project is to analyse how these modulation (joined with coding schemes) schemes affect the performance of a programmes
two user NOMA system. The performance of the NOMA system should then be compared with the Orthogonal Multiple Access (OSA)
systems. The simulations should be carried out in Matlab. Good understanding of the system performance and cellular systems are
required.

Improving Sensor The sensor networks are implemented with different applications in various regions. The sensors can be distributed in regions where Available IoT, Telecommunication
networks life with Energy access for humans is either impossible or dangerous. Access to these kinds of sensor networks, after they are implemented, is very costly and data
harvesting or even impossible. On the other hand, since sensors are general low DSP devices and are in small shape, they only can have small limited communications
fixed batteries. When these batteries run out of life, the sensor does not operate and, eventually, the sensor network will not operate programmes
efficiency.

Energy harvesting, can enhance the battery life of the sensors. If the sensor nodes are equipped with harvesting technologies, they can
have a prolonged life and as a result, the life of the sensor network can be increased. In this project, we aim to design protocols on how
much power should be used from either the fixed battery or the rechargeable battery, so that the life of the networks can be maximized.

It is important to understand the operation cycle of the nodes, as well as the network transmission protocol so that an optimal procedure
can be proposed. The performance of the proposed method should be analyzed, the simulations are carried out mainly in Matlab.

Improving the life of The sensor networks are implemented with different applications in various regions. The sensors can be distributed in regions where Available IoT, Telecommunication
sensors with Energy access for humans is either impossible or dangerous. Access to these kinds of sensor networks, after they are implemented, is very costly and data
harvesting or even impossible. On the other hand, since sensors are general low DSP devices and are in small shape, they only can have small limited communications
fixed batteries. When these batteries run out of life, the sensor does not operate and. programmes

Energy harvesting, can enhance the battery life of the sensors. If the sensor nodes are equipped with harvesting technologies, they can
have a prolonged life. In this project, we aim to design protocols on how much power should be used from either the fixed battery or the
rechargeable battery, so that the life of the sensor nodes can be maximised.

In this project, we focus on single sensor nodes It is important to understand the operation cycle of the nodes. The performance of the
proposed methods should be analysed, the simulations are carried out mainly in Matlab.

Adaptive modulation in One of the major challenges of the future wireless communication networks is the scarcity of available spectrum. There are several Available IoT, Telecommunication
cognitive radio systems proposals on how we can tackle this problem, and cognitive radio technology is one of them. The principle of cognitive radios is that the and data
users with no licence to the channel (referred to by secondary users) can access the spectrum as long as their operation does not communications
harmfully affect the licenced users (referred to by primary users) of the channels. However, there are several challenges to making this programmes
technology a possibility of the future.

This project is about designing an adaptive modulation scheme for the secondary users so that it adapts to the various of the channel. The
purpose of this is to benefit most from the direct channel between the transmitter and the receiver of the secondary user, while at the
same time level of interference that is imposed on the primary users. the proposed method results should be validated using matlab.

Good understanding of the modulation techniques, system performance and probability and statistics is required.
(Delay) Performance This project is to analyse the performance of non-orthogonal multiple access (NOMA) techniques, which is a part of the suggested Available IoT, Telecommunication
analysis of NOMA platform for the 5G systems. Implementing NOMA still faces several challenges, and one of them is that the performance of NOMA in and data
different realistic environment scenarios is still unknown. communications
programmes
The project is to focus on the delay performance analysis of NOMA using BPSK in Rayleigh fading channels. The metrics to be considered
include the buffer overflow probability, the buffer status, the probability of non-empty butter and the source rate. The aim of this project
is to analyse these the performance metrics of a two user NOMA system. The performance of the NOMA system should then be
compared with the Orthogonal Multiple Access (OSA) systems. The simulations should be carried out in Matlab. Good understanding of
the system performance and cellular systems and strong mathematical and knowledge of statistics are required.

Ognibene, Dimitri Big data to predict Predicting other people’s preferences and behaviours is a skill that we use in every social interaction. While this skill is extremely natural Available MSc Advanced Comp.
change of user and common, it indeed has several complexities rooted in the individual intrinsic, experiential and transitory emotional differences. With Sci., MSc Computer
behaviours the advent of social media, but in general with software systems supporting a huge number of users, the challenge of developing systems Engineering, MSc
endowed with similar skills is getting more and more important and more complex considering ethical and legal issues like privacy. It can Intelligent Systems and
be even more important to understand what makes users change their behaviours not only in commercial contexts but especially in Robotics, MSc Computer
educational and political ones. For example, did people opinion toward Brexit changed? What was the cause? Did a review changed Games
buyers attitude toward a top item on Amazon? Why an Amazon review is a review ranked higher than others? Why some users spread a
fake news the first time they see it while others require more exposition and others even criticize it [1]? Do people spread more hate-
speech when they are upset? Do people get addicted to likes and online fights? Did students drop a module after a because they found a
lab too difficult or because the performance in the last test were too low? This project is quite open-ended. An approach could be to
initially find users showing interesting behaviour dynamics (e.g. switching opinion) and gradually adding contextual information to the
analysis (e.g. the user social sphere). To do this, depending on the favourite domain, the student can collect new datasets of user-user and
user system interaction (e.g. amazon, twitter, instagram, facebook and moodle) or use existing ones and explore them using different
state-of-the-art methodologies. Successively, the student will create probabilistic generative models of the users [2,3] and of the network
to study the effects of the different parameters on the generation of such behaviour variations and the temporal evolution of the user’s
beliefs and opinions. Other efforts in this framework comprise the creation of tools that integrate and store observations over multiple
sources (e.g. youtube and twitter) or allow to collect data specifying high level descriptions of the data to be extracted. The specific goal
and details of the project will be disclosed, discussed and agreed between the student and the supervisor. It is possible to be involved in
ongoing research projects.
[1] Wineburg, Sam, et al. "Evaluating information: The cornerstone of civic online reasoning." Stanford Digital Repository. Retrieved
January​ 8 (2016): 2018.
Deep learning for event [2 ] S. Linderman,
Unstructured dataM.
in Johnson,
the formA.
of Miller,
text andR. richer
Adams, D. Blei, and
multimedia L. Paninski,
content “Bayesian
are the Learning
preferred way toand Inference
store in Recurrent
and exchange Switching
information between Available MSc Advanced Comp.
prediction from people. However, this poses different challenges for their usage by automatic systems. The goal of this project is use state of the art deep Sci., MSc Computer
unstructured data learning algorithms to predict real life events from unstructured data. Examples could be elections, football league. This will involve the Engineering, MSc
development of a relation extraction system, which aims to identify the relationship between two nominals (eg.: E1,R,E2 where E1,E2 are Intelligent Systems and
entities and R a link between them), integrated with state of the art NLP systems such as BERT (Devlin et al., 2018), (Bidirectional Robotics, MSc Computer
Encoder Representations from Transformers) and predictors trained the value of interest Games

Deep Reinforcement The diffusion of information on social media is currently based on the the use of reccomendation algorithms that usually are based on Available Not specified
Learning for Social matching current user preference. In several context, e.g. education, high polarization, campaigns, etc.. Greedy matching of user
Network Governance preference with content and other users my not be effective (e.g. backfiring effect). In this project the student will explore the ues of
reinforcement learning algorithms to enable more effective information diffusion on the social media

Detection of toxic Social media content may have deleterious effects on its users (and moderators). Recognising and evaluating such posts can use classical Available Not specified
content and behaviour content classification algorithms but also analysing the reaction of the other users (which may reduce the data demands). In this project
online and /or user the student will explore different state of the art algorithm increative ways to deal, when necessary, with the lack of available datasets
emotional reaction

Thought control for a Projects are available for the application of different algorithms and the development of new ones for the problem of multi robot control Available Not specified
Swarm of robots through brain signals and eye tracking/head movement possibly in an augmented reality setup. The swarm can be endowed with skills like
exploring an area of choice, approaching a target. The user could be able to select the whole swarm or a single robot for control or
receive input. Navigation control can be based on different principled algorithms supported by external collaborators (e.g. Vito Trianni,
ask me for his talk slides). Adaptation to user request and perception can be based on different ML algorithms. The projects will be
performed in collaboration with BCI group and/or external collaborators

Deep Reinforcement Projects are available for the application of different algorithms and the development of new ones for the problem of active robotic Available Not specified
Learning for Single and perception in different contexts (e.g., a swarm of drones inspecting an large field or a single robot verifying the state of an oil platform
Swarm Robot Active parts). Typical problems to face would be the integration of information from multiple observations/robots, the request for high-precision
Vision motion and estimation as well as limited amount of training data. The work can focus on the vision, control and/or generative models
(e.g. GAN) to deal with the issue of limited training data. Interaction with other international prestigious groups would be possible.

Human-Robot Proactive Predicting other people’s preferences and behaviours is a skill that we use in every social interaction. While this skill is extremely natural Available Not specified
Interaction and common, it indeed has several complexities rooted in the individual intrinsic, experiential and transitory differences and even more in
the difficulty of perceiving others’ physical and perceptual state, i.e. the parts of the environment which are relevant for him, such as
affordance and obstacles. In the last years, with the continuous advance of robotics and, in general, vision enabled intelligent systems, the
challenge of developing systems endowed with similar skills is getting more and more important. This skill is crucial for enabling robots to
get out of the factories and help us in our everyday life (homecare and driving assistance), and in conditions of special need, such as
hospitals, building sites, and emergency situations. As an example, imagine drones in a disaster site anticipating first responders’
movements and alerting their human partners of unseen relevant dangers they are about to step on (holes, obstacles, unstable or hot
surfaces) and survivors beyond debris and walls [4]. The project is quite flexible and will focus on the implementation of active social
perception skills [1-3]. It will involve developing and implementing new algorithms for activity learning and recognition as well as sensor
control in simulation [10] and/or translating them to a real robot or drone or even work on datasets [5-9]. We can choose ​different
techniques ​to develop the system from deep learning to symbolic planning passing from bayesian probabilistic modelling. It is possible to
consider ​different sensory modalities ​(vision, audio or both, human partner eye-tracking), output/control modalites (moving robot
camera, flying, interacting with objects, AR feedback to the user). The work can be set in different ​tasks ​such as aiding a human first
responder to explore the site of a disaster [4], alerting someone that is going to bump or stumble, helping a human partner DJ [3], chef [5]
or housekeeper, monitoring operations involving multiple actors (discussions, works in building sites or driving in a trafficked road) and
more. The work can be more theoretic ​oriented​​, e.g. evaluating the gain of using active camera control while recognizing/learning
others’ behaviours, as well as more practical, creating datasets and tools to annotate them [6-8] or transfering to one of the robotics
platforms of the Robotic Arena previously developed architectures [1,4,5]. ​

Perdikis, Serafeim Benchmarking artifact In this signal processing and machine learning project you will benchmark existing methods in brain-computer interaction (BCI) for Available All degrees
removal methods for EEG artifact detection and removal from electroencephalography (EEG) signal and test the ability of independent component analysis (ICA)to
signal and evaluating the identify and remove artifacts without affecting useful signal components. A novel database will be employed, where 12 people have been
efficiency of independent performing common tasks known to inject artifacts into EEG while synchronously recording EEG, facial electromyography (EMG) and
component analysis accelerometers/gyroscopes mounted on the users' heads. This work is expected to lead to a scientific publication.
Assessing awareness of In this project you will assess the existence of motor imagery (MI) correlates in electroencephalography (EEG) signal, in a large database Available All degrees
patients with disorders of of more than 30 patients with disorders of consciousness (DOC). DOC patients were instructed to imagine moving their affected arm, or
consciousness with open- rest. MI trials were followed by functional electrical stimulation of the arm allowing patients to push a buzzer button. You will evaluate
loop motor imagery brain- the extent to which EEG-based neurophysiological metrics correlate with commonly used clinical tests to assess patient awareness and
computer interaction reveal their added value. This work is expected to lead to a scientific publication.
coupled with functional
electrical stimulation

EEG signatures of error In this basic neuroscience project, you will analyze electroencephalography (EEG) data recorded while 25 subjects were performing Available All degrees
processing during ballistic ballistic movements with a computer mouse. The movement target could suddenly deviate and subjects were instructed to attempt to
movements and the correct the movement on the fly. Pre- and post-experimental transcranial direct current stimulation (tDCS) was applied. You will work on
effects of transcranial identifying EEG signatures of human error processing in this task and on revealing any effects of tDCS on the ability of subjects to correct
direct current stimulation movement trajectories. This work is expected to lead to a scientific publication.

Benchmarking signal In this project, you will have the chance to benchmark different signal processing and machine learning methods for feature extraction, Available All degrees
processing and machine feature selection and classification of mental tasks in a big motor imagery brain-computer interface dataset. You may also propose your
learning methods on own methods to be tested in this context. This is a unique dataset of more than 60 able-bodied users and more than 20 users with
motor imagery brain- disabilities, rendering it ideal for method comparisons. The dataset contains both open-loop and closed-loop BCI sessions for each subject.
computer-interface data - This work should ideally lead to a scientific publication.
Co-supervision with Dr
Sebastian Halder

Developing a MATLAB In this project, you will extend an existing MATLAB toolbox for brain-computer interface (BCI) data analysis with new features. Features Available All degrees
toolbox for aggregate will either be developed from scratch by writing original MATLAB code and the corresponding documentation following the toolbox's
analysis of brain- norms, or by wrapping the equivalent functionality from other open-source BCI toolboxes like FieldTrip and EEGlab. New features will
computer interface data correspond to known pre-processing, artifact detection and removal, feature extraction, feature selection or classification methods
employed in BCI. Testing of new features will be performed with EEG data derived by real BCI experiments.

Developing for an open In this project, you will develop for an open-source, modular C/C++ platform for real-time closed-loop brain-computer interface (BCI) Available All degrees
source brain-computer experimentation. Developments may concern new acquisition plugins for supporting additional hardware devices, porting the (currently,
interface experimental mostly Unix-oriented) platform to Windows, porting core analysis infrastructure from MATLAB to Python and/or C++ or adding small
platform useful features. Your work will be tested with real BCI experiments where you will also learn to set up the protocol and execute a BCI
experiment. Previous experience in C/C++ development is highly recommended.

Analyzing the data of a In this project you will extract and study brain plasticity metrics in electroencephalography (EEG) signal, analyzing a large database of 27 Available All degrees
clinical trial assessing the acute stroke hemiplegic patients that have undergone 15 sessions of either a real or a sham brain-computer interface (BCI)-triggered
effects of a brain- functional electrical stimulation treatment for rehabilitation of the upper limb. You will learn to apply advanced EEG processing
computer interface and techniques, statistical analysis for randomized controlled trials and you will search for relations between plasticity indices and clinical
functional electrical outcomes. This work is expected to lead to a high-impact scientific publication. Experience with MATLAB software is desired.
stimulation on upper limb
acute stroke
rehabilitation

Virtual reality feedback In this project you will develop a virtual reality (VR) application in a VR engine of your choice suitable for the Oculus Quest head mount Available All degrees
for a visual attention display device. The application will function as the "feedback" module of a brain-computer interface (BCI) protocol relying on a subject's
brain-computer interface visual attention (VA). VA BCIs decode the neural correlates of covert or overt shifts of attention within a subject's visual field. They can be
protocol used either as an assistive device (i.e., where one can drive a wheelchair left/right by the respective shifts of attention) or as a
rehabilitation tool.

EEG and fNIRs Brain-computer interfaces often employ imagery of motor tasks. Electroencephalography (EEG) signal is the most common option. Available All degrees
coregistration for motor However, it is beneficial to use additional measurement techniques that complement the EEG, like functional near-infrared spectroscopy
imagery brain-computer (fNIRS). In this thesis you will use hardware available in the Essex BCI lab to record a dataset of combined EEG and fNIRS while
interaction - Co- participants are performing mental imagery of motor tasks. Then you will use toolboxes written in Python to analyse the data. Finally, you
supervision with Dr will implement an online filter based on this analysis. This work should ideally lead to a scientific publication.
Sebastian Halder

Rassouli, Borzoo Available to supervise Available


projects.

Raza, Haider Kaggle-based machine Do you aspire to do Machine Learning, Data Science, or Big Data Analytics? Are you ready to start solving real-world problems? I highly Available All degrees
learning competition recommend you take a look at Kaggle Competitions Kaggle is a platform for data science and machine learning competitions. The coders
for machine learning and data science across the globe compete for each other to achieve a higher rank and win prize money The
participants are from industry, universities, and privately. If you feel you can do it, contact me to discuss it further. Let's do it!
https://en.wikipedia.org/wiki/Kaggle
https://www.kaggle.com/

Transfer Learning for The performance of brain-computer interfaces (BCIs) improves with the amount of available training data; the statistical distribution of Available All degrees
handling non-stationarity this data, however, varies across subjects as well as across sessions within individual subjects. This difference in the distribution of data is
in EEG/MEG generally known as the covariate shift in training and testing distributions. Currently, transfer learning techniques in BCIs are used to
exploit shared structure between training data of multiple subjects and/or sessions to increase performance. This project may utilize
transfer learning in the context of BCIs that can be applied to any arbitrary feature space of electroencephalogram (EEG) and
magnetoencephalogram (MEG) features.
https://arxiv.org/pdf/1805.01044 https://arxiv.org/pdf/1512.00296.pdf

Deep Learning to improve One of the major issues in brain-computer interfacing (BCI) is the non-stationarity in the neurophysiological signals. These signals are Available All degrees
the performance of subject-dependent, which means that for a specific user a BCI system has to be built up through extensive training with this user and can
Motor-Imagery Detection work well on this user only. This difference in the distribution of data is generally known as the covariate shift in training and testing
in EEG/MEG-based BCI distributions. The aim of the project will be to use a deep learning approach for learning extracted features and select subject-invariant
features from brain signals (EEG), which would be substantially different from traditional methods for feature extraction in BCI systems.
https://arxiv.org/pdf/1703.05051.pdf https://github.com/robintibor/braindevel

Data Wrangling for PDF It is often the case with data science projects that you’ll have to deal with messy or incomplete data. The raw data we obtain from Available All degrees
data files different data sources is often unusable at the beginning. All the activity that you do on the raw data to make it “clean” enough to input
to your analytical algorithm is called data wrangling or data munging. In today’s work environment, PDF became ubiquitous as a digital
replacement for paper and holds all kind of important business data (reports, invoices, purchase orders, HR forms etc.). But what are the
options if you want to extract data from PDF documents? Manually rekeying PDF data is often the first reflex but fails most of the time for
a variety of reasons. With the amount of PDF data and data sources rapidly growing and expanding, it is getting more and more essential
for the large amounts of available PDF data to be organized for analysis. If you are interested in developing a pipeline for handling PDF
data files to extract key information e.g. text, tables, images, graphical data, this is a chance for you to learn and solve an existing data
science problem.
Reed, Martin J; Securing software defined The future of networking is to use software to control the switches and routers, in particular using something termed "software defined Available All degrees
networking networking." While this has many advantages, it means that modern networks are controlled by a centralised software entity that is both
a single point of failure and open to software related attacks that are common in the operating systems that such a system uses. In this
project you will start with a provided software defined network and you will attack it and devise methods to defend it. The project will use
linux based emulation systems and penetration testing tools such as Metasploit.

Blockchain for securing Blockchain is new distributed authentication/consensus technology which you might have heard about through Bitcoin which uses Available All degrees
network functions Blockchain technology. This project will use Blockchain to authenticate the software components used in a complex system such as a
software controlled network. This project will involve programming an application in a Blockchain system such as Etherium and would
introduce a student to this exciting new technology.
Fog computing Fog computing is set to take cloud computing close to the edge of the network. One of the main purposes of this is to support the Available All degrees
Internet of Things. This project will investigate how computing resources can be controlled and moved to where they are needed. The
project will use linux and the cloud system Openstack.

Mitigating denial of This project will investigate denial of service attacks, and more importantly how to stop or mitigate the attacks. The project can be Available Only available to students
service attacks carried out either through simulation or by using a practical investigation using routers and servers available in a teaching laboratory. taking Computer
Networks and Security

IoT security The Internet of Things (oT) is a fast evolving set of applications for both Industrial and domestic use. However, with the large number of Available All degrees
devices and wide range of usage scenarios there is no one straightforward security solution for IoT. Consequently, there are an increasing
number of IoT systems that exhibit serious security flaws. This project will look at IoT security at both the larger system architecture and
in the application of security in the devices themselves. Whether it concentrates on overall security architectures, or at lower level
security implementations will depend upon the interests of the student undertaking the project.

Security in Cloud based Cloud based systems and applications are becoming highly prevelant. This project will first investigate cloud systems and applications by Available All degrees
applications installing and configuring a cloud based system. It will then go on to investigate user authentication in the cloud and the problems of
federated authentication. Possible solutions will look at aspects such as two-factor authentication. The project will involve using linux
based control platforms and open-source cloud architectures such as CloudStack.

Richerby, David The Moran Process The Moran process is a Markov chain that models the spread of genetic mutations through networks of organisms, and similar situations Available All degrees
such as ideas spreading through social networks. At each step of the process, an organism is chosen with probability proportional to its
"fitness" score, and it reproduces, by replacing a random neighbour in the network with a copy of itself. In the simplest version of the
process, there is one mutation and organisms either have it or don't: organisms with the mutation have one fitness score and organisms
without it have another. In another version of the process, we consider a two-player game (e.g., prisoner's dilemma), which the mutants
play with one strategy and the non-mutants with another. An individual's fitness is its average game payoff against its neighbours.

We are mostly interested in two questions about the process: what is the probability that the mutation spreads to the whole population,
and how long does it take on average? We have good answers to these questions for the simple, fixed-fitness version of the process. The
project is to investigate the game-payoff version. This will be done by producing software that simulates the process, and analyzing its
behaviour on different kinds of networks and with different games. A mathematically strong student may be able to use results from this
study to guide a mathematical analysis.

Constraint-based puzzles Many pencil-and-paper puzzles, such as sudoku, picross/nonograms, kakuro etc. are constraint satisfaction problems. We are given some Available All degrees
kind of grid and we must fill in the entries in the grid -- e.g., with numbers or colours -- subject to constraints such as "Each row must
contain each digit 1-9 exactly once.". There are many algorithms for generic solving constraint systems, and many heuristics for guiding
backtracking searches. The project is to compare the effectiveness of these algorithms and heuristics on these kinds of puzzles, and to
compare them against algorithms designed for the specific puzzles, incorporating techniques that humans use to solve them.

Student-suggested I will consider student-suggested projects in the area of algorithms. Available All degrees
projects

Scherer, Reinhold A walk through the park The aim of the project is to create an immersive 3D experience for people with disabilities or bedridden people to reconnect with nature Available All
360-degree Virtual and other familiar environments. 3-D video footage is provided. A suitable navigation strategy is to be developed and implemented in
Reality (VR) videos Unity for the Vive Pro VR headset. Experiments should be performed to assess user reactions and emotions.

You better not be doing The aim of this project is to design and implement a monitoring solution that reminds people with dementia not to carry out certain Available All
this! actions. For example, to prevent them from putting a kettle on the stove and then forgetting it. When the kettle is detected on the stove
(e.g. by instrumenting objects with RFID) then a video plays (e.g. via Augmented Reality) that reminds the person not to carry out the
action.

Learning disability and The aim of this project is to improve speech recognition systems for persons with speech impediment. This can be achieved by improving Available All
voice recognition for signal processing methods, use of artificial intelligence and machine learning, or by implementing new stand-alone solutions. Project
speech impediment topics will be defined according to the interests of the students.

Complex adaptive A complex adaptive system (CAS) is a system in which many independent elements or agents interact. CAS are dynamic systems able to Available All
systems for brain- adapt in and evolve with a changing environment. The aim of this project is to assess the usefulness of CAS to track changes in
computer interfaces electroencephalogram (EEG) brain signals for brain-computer interface applications.

Body/Muscle-computer A muscle-computer interface is an interaction methodology that directly senses and decodes human muscular activity and does not rely Available All
interface on physical device actuation. The goal of this project is to implement a creative solution for a specific HCI scenario (e.g. hand gesture
detection for interactive systems). Project topics will be defined according to the interests of the students.

Narrative game for brain- In order to use a brain-computer interface a user must train for longer periods of time. However, modern training paradigms are boring Available All
computer interface and have poor graphic design (e.g. lines and arrows on a black screen). The goal of this project is to design and implement a narrative
training game that is simple, engaging and requires only a binary input (e.g. based on the idea behind The Stanley Parable game).

MyProject (student It is open to students to suggest topics in the areas of brain-computer interfacing, neural engineering, adaptive statistical signal Available All
suggested projects) processing, (neuro)rehabilitation and assistive technologies. Please make an appointment with Prof. Reinhold Scherer to discuss
possibilities.
Scherp, Ansgar Multi-label Graph CNN for Goal of this thesis is to combine to very successful and interesting areas of machine learning / Deep Learning on text. Graph CNN, or Available All
Few-Shot Text Graph Convolutional Networks, for text classification have a simple but very effective approach for text classification. Few-short learning
Classification can deal with limited training data. Many documents have more than one label, thus defining the task of a multi-labelling of documents.
This topic has the chance, if successful, to be submitted and accepted at an international scientific conference.

Online History Tracking Vocabularies of the Semantic Web are subject to changes. Thus the data that uses of vocabularies' terms needs to be updated. Until now, Available All
System for Vocabularies no system exist for tracking the vocabularies changes and gives a full statistics about adoption and usage in terms of Pay-Level-Domains
(PLDs). This Work shall analyze Semantic Web vocabularies’ changes, and give detailed statistics for terms’ usage and adoption, by
parsing the most known datasets (such as DyLDO, BTC, dbpedia, WikiData,…). The system should show what is changed, how it is used,
and which PLDs adopted new coined vocabularies’ classes and properties based on the data crawled by a well-known datasets.

Linked Open Data Linked Open Data is a very popular way to publish and interlink data on the web. It is support among others by media industries, libraries, Available All
Crawler governmental organization as well as commercial corporates. For the example, since February 2017 the EU Open Data Portal publishes
all metadata of the EU member states in Linked Open Data. Goal of this work is to develop a light-weight crawler for conducting
continuous crawls of Linked Open Data. This follows ideas of existing data collection efforts such as DyLDO.

Flexible visualization of The Debeka group, a large German insurer, is developing a semantic information system that will handle all the artifacts of their IT Available All
linked data Landscape in relation to each other. Examples of such artifacts are functional requirements, business process models or software
modules. The data of the information system is stored as linked data and made available to various display systems. Semantic data
processing uses ontologies that take a very technical view of IT artifacts. Since the data of the information system is difficult for specialist
analysts to access, a simpler view of them is needed. The aim of this work is the development of a framework for the flexible visualization
of linked data. The visualizations should be adaptable to individual needs and abstract from the technical modeling details of the
ontologies. They are intended to enable a user to prepare the information relevant only to him at a level of detail that is appropriate for
him.

Extreme Multi-label Text I would like to discuss your original ideas and advise you on the problem of Extreme Multi-label Text Classification (XMLC). Based on own Available All
Classification prior work in the field, we will identify a pressing current issue in the field, define it and outline an agenda. To get an introduction to
XMLC, please have a look at: 1) Quadflor: a text-processing pipeline for XMLC of documents and its evaluation. Given a domain-specific
thesaurus with descriptor labels, different algorithms learn how to assign labels to documents from a training set. The framework
supports concept extraction, synonym set resolution, spreading activation including hierarchical re-weighting. Quadflor documentation
(lucidML): https://github.com/quadflor/Quadflor/blob/master/Documentation/technical_documentation.pdf Code for the JCDL paper
"Using Deep Learning for Title-Based Semantic Subject Indexing to Reach Competitive Performance to Full-Text"
https://github.com/florianmai/Quadflor 2) Workshop on "Extreme Multilabel Classification for Social Media", April 23, 2018, Lyon, France
Website: https://sites.google.com/view/xmlc/

Extraction of Metadata Approaches for extracting metadata from scientific publications mostly concentrate on detecting standard bibliographic information such Available All
from Documents at the as title, author, venue, and publication year. There are also some works that aim at extracting structured data from the publications such
Example of Statistical as data published in tables. However, further details about the publications are often not extracted such as statistical data. In this thesis,
Metadata we aim at developing a rule-based and a machine learning approach for extracting statistical metadata in scientific papers. The rules are
based on standards how statistical results should be published in scientific papers. Subsequently, they are used to induce training data for
a machine learning model such as Deep Convolutional Neural Network. For experiments, large-scale datasets of scientific publications will
be provided.

Semantic Analysis of Text I would like to discuss your original ideas and advise you on a topic related to Text Analytics. The topic can be on one of the various tasks Available All
in natural language processing and text analytics like extracting structure from text, improving the semantic understanding of language,
game-based approach to text analytics etc as long as some connection to the semantic analysis of text is provided.

Business Process This project will examine the benefits and challenges of using deep learning to provide predictions on the evolution of graph structured Available All
Predictions using Graph data. The specific use-case for these graph structures is the recording of temporal relationships in business process executions.
Structures in Deep Predictions can take the form of the next task that will occur, the time of the next task occurrence, the predicted final outcome of the
Learning (industrial business process, or some combination of these. Points of comparison with other predictive approaches can be provided.
project with BT, co-
supervised by Dr. Paul Skill requirements
Taylor) The student will need some knowledge of python, knowledge of tensorflow would be a plus but is not essential.

Refs:
- https://github.com/deepmind/graph_nets
- https://rd.springer.com/chapter/10.1007/978-3-319-71078-5_24

Singh, Amit K Improving Battery Smart phones typically consume energy supplied by a battery and their energy consumptions need to be optimized to reduce energy costs Available Any degree
Lifetime and Age of Smart or increase battery lifetime. Further, they should operate at low temperature to have a long age before they die and perform reliable
Phones operations. Both design-time and run-time optimizations are required for energy- and temperature-efficient operations. At design-time,
the efficient designs can be explored to start the operations and run-time adjustments need to be made due to changing environments
such as increase/decrease in the device usage and downloading a new application to the device.
A multi-core processor used in modern smartphones will be considered to carry out the project.

Sperm quality The ability to accurately and precisely measure sperm quality is of great importance in the areas of agriculture, reproductive toxicology, Available Any degree
measurement by machine evolutionary biology, and, of course, human fertility. Infertility is a multifactorial problem affecting 1 in 7 couples in the UK, with male
learning enabled shape infertility or subfertility being a contributing cause in approximately half these cases. Clinics routinely screen patient semen samples for
and texture analysis of metrics such as shape, motility and structure, but even the best-automated systems are limited in the resolution of data they can capture,
images due to constraints on imaging time and expertise.
This project will develop convolutional neural networks that can better classify classes of abnormal sperm integrating shape, texture
and/or structural information from images. Along with our existing sperm samples from mouse strains, we have access to pig semen
samples from an international pig breeding company, and are extending our work into humans in collaboration with an infertility clinic.
There will be guidance from an academic from Life Sciences as well, relevant to the biological aspects.

Attacking and Securing The attackers can access the confidential data if secure execution is not performed. Since these data can be sensitive, it can be Available Any degree
Embedded Multi-core concerning. Therefore, companies have started manufacturing chips to process the sensitive data in a secure zone. However, it is still
Systems attackable as demonstrated in the literature.
This project plans to examine some of these popular methods to attack attach and then explore and analyse various methodologies that
can be employed for secure execution of programs on processor cores. The analysis will involve to find a lightweight methodology for
performing secure execution during run-time.
The successful handling of the project will require some knowledge of processors and C/C++ programming.
An advanced multi-core processor will be considered to carry out the project.

Profit and Energy Large scale HPC data centers integrate several many-core architectures to enhance their processing capability, but a huge energy is Available Any degree
Optimizing Resource required to operate them, around 1.5% of the worldwide electricity consumption and it is expected to increase rapidly in future. The
Management for Many- existing resource allocation approaches assign resources to jobs as soon as jobs arrive or resources become available for the queued jobs.
core HPC Data Centers However, such immediate decisions might provide allocations on lower number of cores leading to high completion time and a higher
number of cores might be available immediately after taking the decision. Therefore, it might be beneficial to look the status of busy
cores to know if they will be freed soon and use them as well to allocate the job by delaying the allocation instance. This project aims to
explore such advantages/disadvantages in terms of profit and energy by changing the allocation instance in the hope that a better
resource allocation can be achieved.
A cluster of Many-Core processors is available to carry out the project
Embedded AI: Employing Embedded AI include software and small devices used by everyday people, with integrated machine/deep learning designed to simplify Available Any degree
Machine Learning in the user’s day-to-day tasks. These tasks could be image classification, object detection, traffic prediction, etc.
Smart-phones This project aims to implement an efficient machine/deep learning method in a multi-core processor used in modern smartphones. The
convolutional neural network or another kinds of learning can be employed. While employing these learning methods, several
optimization metrics can be considered, e.g. performance, energy consumption, reliability, etc. However, the plan would be to consider
only the most important metrics depending upon the kinds of jobs to be performed by employing the embedded AI.
An advanced multi-core processor will be considered to carry out the project.

Quantifying cellular Microscopy is a cornerstone of modern biomedical research, as it allows us to study cells - the smallest units of life. Microscopy is also Available Any degree
behaviour using inherently cross-disciplinary, requiring engineering, optics, computer control and computer analysis. Different types of cells show a wide
automated experimental variety of responses and behaviour in different experimental conditions.
control and image
analysis In this project, we plan to explore such effects by automatically controlling cellular exposure to different fluids on the microliter scale, i.e.
microfluidics. It will require the development of a microcontroller-based mechanism to steer microfluidics on a microscope. By doing so,
the experimental conditions can be varied, and the cells' subsequent response quantified using automated image analysis. This in turn can
provide insight into environmental or biomedical problems, e.g. how some single-celled organisms react to chemical pollution, or how
some cancer cells react to a particular drug. The digital images can then be analysed and quantified using (semi-)automatic algorithms
and pipelines in open-source software.

The project will be co-supervised by an academic from the School of Life Sciences.

Detection of Different Lung cancer is quite epidemic and common among all the types of cancer found in men and women. In fact 13% of all cancer diagnoses Available Any degree
Stages of Lungs Cancer in consists of some form of lung cancer. Recently visual convolutional neural network-based machine learning algorithms have been really
CT-Scan Images Using successful to detect different stages on lung cancer from medical images but given the time and energy consumption of such algorithms
CNN on Embedded due to their computation intensive architecture it is infeasible to train such algorithms on mobile embedded platforms because of
Systems constrained resources on board. Although it could be argued that training could be performed on powerful computing systems and the
prediction is performed on embedded devices but due to the varying nature of such medical diagnoses from region to region and
inaccessibility of internet connectivity in developing countries it is more useful to train the CNN on a mobile embedded platform instead.
This project focuses to design and develop a light-weight CNN based algorithm which could detect different stages of lung cancer on a
mobile embedded systems keeping performance, energy-efficiency and reduce operating temperature in mind. The developed CNN
algorithm needs to achieve at least 90% accuracy in detecting lung cancers with any given dataset on our heterogeneous multi-core
system-on-a-chip platform utilizing both CPU and GPU.

Dataset Used: Armato III, Samuel G., et al. "The lung image database consortium (LIDC) and image database resource initiative (IDRI): a
completed reference database of lung nodules on CT scans." Medical physics 38.2 (2011): 915-931.

Unified control of a Studios, theatres, stadiums, airports etc. all deploy complex audio systems usually with multiple consoles or mixers for control and Available Any degree
networked audio system monitoring. A new open standard for controlling audio systems makes it possible to create a unified or virtual control surface
incorporating networked products such as digital signal processors (DSP) as well as microphones, amplifiers and loud speakers.
This project will demonstrate the potential of unified control by automatically optimising the acoustics of a venue and continuously
monitoring it from anywhere on the network. To do this will require remote control of the embedded algorithms in both the microphone
and loud speaker signal chains. Automatic optimisation can be performed by employing suitable machine learning algorithms as well.
This project is an industrial project with Profusion plc who are expert in audio systems and there will be guidance from an industrial
supervisor as well.

Interactive scheduling BT has developed a field force scheduling system which support automated, manual and hybrid task-resource allocation where field or Available Any degree
(industrial project with desk agent are guided by the system into the set of tasks they are advised to allocate next.
BT, co-supervised by Dr. Such system is interactive and to ensure a smooth adoption of the guidance generated by the system this is important to provide some
Anne Liret) explanation to the user (why does this action or this task is suggested as the best next one- how does it improve the overall schedule of
tasks or the overall productivity of the field agents or the service delivery, as example).
The project proposed to build a model that generates some explanations about the recommended next set of task allocations that the
scheduling system currently produces. This may be done using AI techniques or by simply simulating what alternatives actions would
produce and compare with the best option. This could as well be interleaved with a scheduling model is seen as suitable during the
project.

Skill requirements
Programming in java, some basics in operational research or scheduling is a plus.
Some knowledge in machine learning, CBR, fuzzy logic or other techniques of AI that may be used to generate explanation is a plus but
not required.

Su, Jinya Machine learning in In this practical project, the student will apply state-of-the-art machine learning algorithms to addressing the global grand challenges Available Prerequisites and
precision agriculture faced in modern agriculture. The particular problem of interest is to classify the multispectral aerial image (MSI) into healthy wheat and requirements: MSc in
diseased wheat (infected by yellow rust disease), where the MSI is obtained by an Unmanned Aerial Vehicle (UAV) at an altitude of 20m in Artificial Intelligence,
2019. This problem is also termed semantic segmentation in computer vision community. The student should be familiar with the MSc in Robotics and
foundations of machine learning or AI and is also confident in using Matlab, Python or other languages. Reference: Jinya Su et al. "Wheat Intelligent Systems, MSc
yellow rust monitoring by learning from multispectral UAV aerial imagery." Computers and electronics in agriculture 155 (2018): 157- Big Data and Text
166. If you have any question about this project contact j.su@essex.ac.uk Analytics, MSc in Data
Science. Excellent
programming skills
(Matlab, Python), interest
and knowledge on
applications of machine
learning and AI in real-
world problems.

Deep learning in precision This is a practical project, in this project, the student will apply deep learning algorithms to addressing the global grand challenges faced Available Prerequisites and
agriculture in modern agriculture. The particular problem of interest is to classify the multispectral aerial image (MSI) into healthy wheat and requirements: MSc in
diseased wheat, where the MSI is obtained by an Unmanned Aerial Vehicle (UAV) at an altitude of 20m in 2019. This problem is also Artificial Intelligence,
termed semantic segmentation in computer vision community. The student should be familiar with the foundations of machine learning MSc in Robotics and
or AI, has an interest in deep learning and is also confident in programing using Matlab, Python or other languages. Reference: Huang et Intelligent Systems, MSc
al "A fully convolutional network for weed mapping of unmanned aerial vehicle (UAV) imagery." PloS one 13, no. 4 (2018): Big Data and Text
e0196302.https://uk.mathworks.com/help/vision/examples/multispectral-semantic-segmentation-using-deep-learning.html. If you have Analytics, MSc in Data
any question about this project contact j.su@essex.ac.uk Science. Excellent
programming skills
(Matlab, Python), interest
and knowledge on
applications of machine
learning and AI in real-
world problems.
Deep learning for pest Object detection algorithms can find a wide range of real-life applications. In this practical project, the student will apply object detection Available Prerequisites and
detection algorithms for pest detection (e.g. grasshopper detection). The student can customize the methodological plan according to what he requirements: MSc in
aims to achieve (the long-term aim is a software on mobile phone with robust performance). The student will be interacting with deep Artificial Intelligence,
learning architectures such as Convolutional Neural Networks (CNN), ROLO/YOLO or other architectures. For this project, the student MSc in Robotics and
should be familiar with the foundations of machine learning and AI, has an interest in deep learning and is also confident in programing by Intelligent Systems, MSc
using Matlab, Python or other languages. Reference: Zhao et al “object detection with deep learning: a review.”, IEEE Transactions on Big Data and Text
neural networks and learning systems, 2019. https://uk.mathworks.com/discovery/object-detection.htmlIf you have any question about Analytics, MSc in Data
this project contact j.su@essex.ac.uk If you have any question about this project, please contact j.su@essex.ac.uk Science. Excellent
programming skills
(Matlab, Python), interest
and knowledge on
applications of machine
learning and AI in real-
world problems.

AI in viticulture (disease AI is finding applications in a number of semantic applications. This practical project is to apply deep learning algorithms for the disease Available Prerequisites and
monitoring) monitoring of grapevine yellows symptoms (leaf level disease monitoring). Color images will be used in this project. Various deep learning requirements: MSc in
architectures will be learnt and tested including AlexNet, GoogLeNet, Inceptions, etc. The algorithms can also be applied to other tasks in Artificial Intelligence,
agriculture and horticulture. For this project, the student should be familiar with the foundations of machine learning and AI, and is also MSc in Robotics and
confident in programing by using Matlab, Python or other languages. Reference: Cruz, Albert, Yiannis Ampatzidis, Roberto Pierro, Alberto Intelligent Systems, MSc
Materazzi, Alessandra Panattoni, Luigi De Bellis, and Andrea Luvisi. "Detection of grapevine yellows symptoms in Vitis vinifera L. with Big Data and Text
artificial intelligence." Computers and electronics in agriculture 157 (2019): 63-76. If you have any question about this project, please Analytics, MSc in Data
contact j.su@essex.ac.uk Science. Excellent
programming skills
(Matlab, Python), interest
and knowledge on
applications of machine
learning and AI in real-
world problems.

Automatic drivers’ state Drivers’ state (i.e. normal, drowsy and aggressive driving) can be reflected by their driving performance. For example, aggressive driving Available Prerequisites and
recognition by machine usually involves abrupt acceleration or deceleration. In this project, machine learning algorithms will be applied to recognize drivers’ requirements: MSc in
learning algorithms state automatically for the purpose of safe driving. In particular, features derived from time-series data will be exploited against point Artificial Intelligence,
data. The real-world UAH-DriveSet dataset (https://ieeexplore.ieee.org/document/7795584) will be adopted to assess the performance MSc in Robotics and
of the algorithms. For this project, the student should be familiar with the foundations of machine learning and AI, and is also confident in Intelligent Systems, MSc
programing by using Matlab, Python or other languages.Reference: Dewei Yi, Jinya Su, Cunjia Liu, Mohammed Quddus, and Wen-Hua Big Data and Text
Chen. "A machine learning based personalized system for driving state recognition." Transportation Research Part C: Emerging Analytics, MSc in Data
Technologies 105 (2019): 241-261.If you have any question about this project, please contact j.su@essex.ac.uk Science. Excellent
programming skills
(Matlab, Python), interest
and knowledge on
applications of machine
learning and AI in real-
world problems.

Welcome a student Please make an appointment to discuss possibilities via j.su@essex.ac.uk Available Prerequisites and
suggestion on a project in requirements: MSc in
the area of AI for Artificial Intelligence,
precision agriculture or MSc in Robotics and
other applications. Intelligent Systems, MSc
Big Data and Text
Analytics, MSc in Data
Science. Excellent
programming skills
(Matlab, Python), interest
and knowledge on
applications of machine
learning and AI in real-
world problems.

Thakur, Manoj Available to supervise Electronic Engineering projects Available


projects in 2019-20

Varasteh, Morteza Zero-delay joint source- In this project, the zero-delay source-channel coding problem, and specifically the problem of obtaining the vector transformations that Available Matlab, Communications
channel coding optimally map between the m-dimensional source space and k-dimensional channel space, under a given transmission power constraint systems, Optimization
and for the mean square error distortion will be studied. The following topics will be investigated in depth: 1) The functional properties of
the cost and the necessary conditions for the optimality of the encoder and decoder mappings. 2) An optimization algorithm that imposes
these conditions iteratively, in conjunction with the noisy channel relaxation method to mitigate poor local minima. 3) Numerical results
should be run accordingly and interpreted based on theory.

Wireless information- Radio-Frequency (RF) waves can be utilized for transmission of both information and power simultaneously. The ambient energy sources Available MATLAB,
power transfer on make energy harvesting unreliable for applications where a constant flow of energy is of significant importance. An energy harvesting Communications systems,
frequency-selective technology that overcomes this limitation is Wireless Power Transfer (WPT), where dedicated signals are designed to transmit the Optimization
channels required power at the user end. The main challenge in WPT is to increase the Direct-Current (DC) power at the output of the harvester,
known as rectenna. A rectenna is made of a nonlinear device followed by a lowpass filter to extract a DC power out of an RF input signal.
The amount of DC power collected is a function of the power and shape of the RF signal. In this project the aim is to tackle the design of
channel input for Simultaneous Wireless Information-Power Transmission (SWIPT) under nonlinear constraints. To that end, SWIPT in a
multi-carrier setting subject to nonlinearities of the EH will be studied.

Learning Communications End-to-end learning of communications systems is a fascinating novel concept whose goal is to learn full transmitter and receiver Available Reinforcement Learning,
Systems Without a implementations which are optimized for a specific performance metric and channel model. This can be achieved by representing Machine Learning,
Channel Model transmitter and receiver as neural networks (NNs) and by interpreting the whole system as an autoencoder, which can be trained in a Python, MATLAB,
supervised manner using stochastic gradient descent (SGD). Although a theoretically very appealing idea, its biggest drawback hindering communications systems
practical implementation is that a channel model or, more precisely, the gradient of the instantaneous channel transfer function, must be
known. For an actual system, this is hardly the case since the channel is generally a black box for which only inputs and outputs can be
observed. In this project, it is investigated 1) if methods from the field of (deep) reinforcement learning (RL) can be used to circumvent
the problem of a missing channel gradient. 2) show that knowledge of the channel model and the instantaneous channel transfer function
is indeed not needed.

Machine Learning at the Many emerging technologies involve massive amounts of data collection, and collaborative intelligence that can process and make sense Available Communications systems,
Wireless Edge: of this data. Internet of things (IoT), autonomous driving, or extended reality (XR) technologies are prime examples, where data from Machine Learning,
Distributed Stochastic sensors must be continuously collected, communicated, and processed to make inferences about the state of a system, or predictions MATLAB, python
Gradient Descent Over- about its future states. Many specialized machine learning (ML) algorithms are being developed tailored for various types of sensor data;
the-Air however, the current trend focuses on centralized algorithms, where a powerful learning algorithm, often a neural network, is trained on
a massive dataset. While this inherently assumes the availability of data at a central processor, in the case of wireless edge devices,
transmitting the collected data t o a central processor in a reliable manner may be too costly in terms of energy and bandwidth, and
undesirable due to privacy concerns. Communication is typically more costly compared to processing; therefore, a much more desirable
and practically viable alternative is to develop distributed ML techniques that can exploit the local processing capabilities of edge devices,
requiring limited communications. In this project, we consider ML at the wireless network edge, where distributed processors with local
data samples and connected to a central parameter server (PS) through a shared wireless medium, jointly train a learning model.
Learning to Communicate The explosion of the number of low-power devices in the next decades calls for a re-thinking of wireless network design, namely, unifying Available Machine Learning,
and Energize: Modulation wireless transmission of information and power so as to make the best use of the RF spectrum, radiation, and infrastructure for the dual Python, MATLAB,
Designs for Wireless purpose of communicating and energizing. In this project, the following topics will be investigated: 1) learning a parametric model of a Communications systems,
Information-Power practical energy harvester, accounting for various sources of nonlinearities. 2 learning problem of modulation design for Simultaneous Optimization
Transmission Wireless Information-Power Transmission (SWIPT) over a point-to-point link. 3) an algorithmic SWIPT modulation design inspired by the
results of the learning approach.

Deep Learning for the In communications, we can design transmit signals that enable straightforward analytic algorithms for symbol detection for a variety of Available Machine Learning,
Physical Layer channel and system models (e.g., detection of a constellation symbol in additive white Gaussian noise (AWGN)). Thus, as long as such Python, MATLAB,
models sufficiently capture real effects, we do not expect DL to yield significant improvements on the physical layer. Nevertheless, the DL Communications systems
applications which we explore in this project are a useful and insightful way of fundamentally rethinking the communications system
design problem. In this project, we investigate 1) possibility of learning full transmitter and receiver implementations for a given channel
model which are optimized for a chosen loss function. The key idea here is to represent transmitter, channel, and receiver as one deep
neural network (NN) that can be trained as an autoencoder. 2) Extending this concept to an adversarial network of multiple transmitter-
receiver pairs competing for capacity.

Walker, Stuart D; Options for 40-100GbE Recent advances in mm-wave technology have allowed 10GbE links to operate over 10km. 40-100 Gbe capacity links are a natural Available All degree schemes
mm-wave wireless extension of such performance. This study will examine and plan these ultra-high capacity links.

Use of mm-waves for As a 10Gbe copper-media backbone for trains is now available, this project will consider using mm-wave wireless to provide train-to Available All degree schemes
train communications track side connectivity . It will be a theoretical study.

Railway Data Ail operators should consider how to provide break-free wireless communication to trains. Line-of-sight optical links look interesting Available All degree schemes
Communications using but need careful study..hence this project.
optical links

Statistical averaging This study will review the dynamics of modern and classical music from a statistical point of view. This wall provide a means of Available All degree schemes
techniques for >1000W assessing whether the latest PoE technology will offer stadium levels of sound power using small generators or even solar power.
PoE loudspeaker systems

5G in the home The emergence of 5G communications leaves the question of how this technology would migrate into the homes. This project will look Available All degree schemes
at possible scenarios

Student’s own It is open to students to suggest topics to Prof. Walker provided he can supervise them appropriately. Please make an appointment to Available All degree schemes
suggestions for projects discuss possibilities

Walton-Rivers, Available to supervise Available


Joseph projects

Woods, John C; VIRTUAL With mobile phones being ubiquitous, high definition audio and video now available on handsets, it must surely be possible for people to Available All
TELEPORTATION experience 'virtual teleportation'. This project will use a headset to allow a user to experience the audio and video as perceived by the
other party in the conversation and as such teleport them to another environment. With the addition of a pan and tilt camera the
experience will be enhanced. Users will be subjectively analysed to examine the level or immersion they experience.

BALANCING ROBOT: THE A classic problem in computing, engineering and mathematics: in the same way as it is possible to balance a stick placed in the palm of Available All
INVERTED PENDULUM the hand, it is possible to defy gravity and build a machine to maintain an inverted pendulum in the upright position. The project will see
PROBLEM the design of a small robot with a video camera placed on it to evaluate the effectiveness of a balanced camera for studio filming

THE AUTONOMOUS Sailing boats represent possibly the closest thing we know to a zero emission vehicle. This study would examine the control algorithms Available All
SAILING BOAT necessary to automatically control a sailing yacht, including the rules of sailing, dead reckoning and performing missions. Power
consumption issues and the means by which onboard power could be maintained should also be considered. The study is likely to be in the
form of a simulation using OpenGL or similar.

BAD TIME ENERGY Many domestic homes now have solar panels on the roof which are generating appreciable amounts of power. The energy generated is Available All
SOLUTIONS fed into the mains using a grid tie inverter and the homeowner gets paid according to the number of kilowatts they generate. The
homeowner is paid for the amount they generate, even if they use it all themselves and feed none of it back into the grid. Therefore what
is required is a control system to monitor the rate of generation and consumption and 'dump' any excess into a consumer load such as an
immersion, thereby legally taking energy that would otherwise supplement the grid!

NON INVASIVE HOME Home automation equipment is available on the market but tends to be proprietary and expensive. Most homes nowadays have a Available All
AUTOMATION/DOMOTIC wireless WIFI 802.11 router which could be used to relay home automation information. Up until recently WiFi enabled end points have
S been expensive and power hungry but now with products appearing for the Internet of Things, Low power, low cost development boards
are cheap (£2) and commonplace. Using a wireless router and a collection of ESP8266 chipsets this project will see the design and
construction of a non-invasive home automation system.

A HEAD DRIVEN HUMAN Interaction with a computer is a task traditionally done with a keyboard and mouse and has been with us for around 30 years. However Available All
COMPUTER INTERFACE this is extremely invasive and requires complete commitment from the user, engaging eyes, head, arms and hands. This project is about
controlling the computer using nothing more than a webcam and the motion of the head. The webcam tracks the translation and
rotation of the human head, and images a stick is connected to the end of the user's nose, at the end of the stick is the cursor, thereby
leaving the user hands free and able to control the computer with head movement alone.

The Turing challenge- Machines that can answer questions posed by a user as realistically as if a human were answering have achieved artificial intelligence; Available All
Chatterbots according to Allan Turing. The purpose of this project is to write a piece of software to respond to keyboard input or voice recognition in
an intelligent human way. Particular focus should be applied to following the context of a conversation. The system could be used to
control a personal computer.

WiTricity; Wireless Power By employing air wound coils at resonance it is possible to transfer energy across free space efficiently up to two coils distant. Applications Available Electronics
Transfer include phone charging, domestic appliances and wireless computing devices. This project will examine the implementation of a practical
system with dissimilar sized transmission and reception coils and single transmitter, multiple receiver set-ups for use with a ‘charging
table’.

THE IMMORTAL A machine which has access to ambient energy in the form of RF, solar etc can be made to do useful work. If the work is powering a Available All - but Electronics
MACHINE microcontroller, then a system that can power itself and keep itself alive for extended periods even in the presence of an preferred
interrupted/intermittent energy source is The Immortal Machine; the machine that lives forever.
Teaching a machine to Using a watershed transform or K-means an image is partitioned into a number of small homogenous regions which are recorded into a Available All, Good coding skills
see tree based structure. The database can be indexed using a software query language to obtain regions of the image and the addition of a
standard syntax can form a generic way in which machines can interact with images without understanding the underlying mechanisms.

Powering a load from an If a source does not have sufficient power to drive a load, common sense dictates that no useful work can be done. However by storing Available All
inadequate source energy into a low loss capacitor and analysing the charging state using a microcontroller, once sufficient energy has been stored it can be
released for a short period of time into the load to perform the required useful work. Applications include pond pumps etc.

VR modelling from still With the advent of high quality VR environments such as the Occulus Rift and the HTC vive, the potential to map 3D environments for a Available All but preferably CS and
images number of reasons such as flood analysis exists. This project will explore the smooth pipelining of still image acquisition to the games
development of a fully interactive room scale VR experience for engineers and managers of equipment and infrastructure
Magnet deceleration A strong rare earth permanent magnet will induce an equal and opposite magnetic field in a moving conductor. This research will Available All
examine the deceleration of a magnet on the end of a pendulum as it approaches a sheet of copper. Counterintuitively the magnetic
pendulum when raised will not strike the copper sheet. Of interest is the wireless measurement of the acceleration and deceleration of
the pendulum as it approaches the conductor. Computer modelling, interface and graphics are encouraged. The project is open to MSc
students of all disciplines.

Students own suggestion It is open to students to suggest topics to Dr Woods provided he can supervise them appropriately. Please make an appointment to Available All
for projects discuss possibilities.

An on-line examination Design and build an online examination module – allowing the individuals running the training programs to adapt and change the Available Prerequisites: Essential:
module for Ipswich-based questions ‘on the fly’ to suit the changing needs of the agency and industry. Web programming
company, Jacob Bailey • Link ‘examination results’ to unique users, and automatically update JacobBailey intranet profiles to reflect status of individuals
(Advertising Agency) qualification.
•Contact project supervisor for further details and to arrange an initial meeting with JacobBailey
• Develop ‘gamification’ linked to user profiles on JacobBailey intranet to demonstrate status of users through The Jacob Bailey Academy
(Undergraduate, Graduate, Bachelor or Master) and showcase all internal and external qualifications and areas of expertise
Here is the full list of programming languages the company currently uses, into which your coding would have to integrate: Javascript,
PHP, HTML, CSS, VueJS, Laravel

Yang, Kun Firmware Update for ZigBee (IEEE 802.15.4) is widely used in sensors/actuators for data communications due to its nice feature of very low energy Available Student of any degree
Zigbee Wireless Nodes consumption. Sometimes there is a need for Zigbee firmware to be updated. The aim of this project is to design a Zigbee firmware scheme but better for
update mechanism and to implement on off-the-shelf Zigbee hardware. students with embedded
system skills
Message Queuing In the world of Internet of Things (IoT), small objects (things) need to talk to each other to fulfil a coordinated task together. Message Available Student of any degree
Mechanism for queuing can be an effective means to this. This project is to design and prototype a message queuing mechanism so that small things can scheme but with good
Communication between exchange messages. programming skills
Small Things
A Smart VoIP Gateway This project aims to bring two exciting fields, namely, VoIP (voice over IP) and WiFi, together into one piece of device. The project will Available Student of any degree
over Open Source WiFi bring an open source VoIP server such as Kamailio on top of another widely-used Linux implementation of WiFi platform called openWRT. scheme
A VoIP demo is expected in addition to performance evaluation of the system.
Cloud Computing Cloud computing has become an essential enabling technology for the everyday operation of many companies and it is getting Available Student of any degree
Monitoring System increasingly more popular across IT fields. The project is to install, configure and to implement a monitoring system for an existing cloud scheme but with strong
platform called OpenStack, which is widely advocated by industries. This project will provide the candidate student a good opportunity to Linux background.
study cloud computing in a practical fashion.

Mobile Edge Computing With increasingly more end devices connecting to the Internet and requiring processing sometimes as driven by the popularity of the Available Student of any degree
Internet of Things (IoT), there is a need to push computation towards edge, i.e., closer to end devices. This project is to look into this scheme but with strong
exciting new paradigm that is welcomed by industries and implement a case study where a mobile edge computing (MEC) system is programming skills
prototyped and demonstrated by implementing a simple use case, such as face recognition on a Rasperry Pi system.

Communication and Computation-intensive tasks that are difficult to be executed on mobile devices can sometimes be offloaded to remote cloud to execute Available Student of any degree
Computation Cooperation and then the results are returned back to the mobile devices. This remote execution will incur extra communications. How to strike the scheme but with good
(3C) balance between these two resources: communications and computation, is the focus of the project. This project involves theoretical mathematical
modelling and optimization. background

Zakhleniuk, Nickolay A Simulation-Based Study In this project the student will develop, design and investigate operation of long-haul realistic optical fibre network. One possible option is Available - can MSc Telecommunications
A; and Performance Analysis to design the telecommunication network for your native country, evaluating various aspects of such network (like the cost of the project, be carried out & Information Systems
of Long-Haul optimal architecture, reliability, the required operation support and services, etc.). Since the introduction of optical switches can reduce by 2 students
Telecommunication the amount of expensive and power consuming electrical equipment in a network, the project will also include the analysis of the inclusion
Networks of optical switching technologies from both a physical performance and financial aspect. The operation of the network will be simulated
using the VPI (Virtual Photonics Inc.) software tool. The main attention in the modelling will be paid to investigation of various network
components and their parameters on the overall network operation. Good knowledge of basics of photonic networks and devices is
required. No particular computer programming skills or software code writing expertise is required.

Statistical Analysis and A network dimensioning is based on three key points, namely – the performance requirements, the available bandwidth and the offered Available - can MSc Telecommunications
Equivalent Bandwidth load. Bandwidth provisioning is an important parameter of Quality of Service since other parameters depend on the bandwidth be carried out & Information Systems
Estimation in IP Networks provisioned in the link. A link in a network is provisioned and managed by the network operators based on standard rules of thumb which by 2 students
Using Real Network generally tend to over-provision the link. This is because the measurements taken for determining the rules of thumb are coarse and
Traffic Data generally taken at longer time intervals. The need of the hour is to obtain a mathematical formula or the other appropriate means (tools)
which help in dimensioning the network with parameters measured at smaller time intervals of a few milliseconds to one second since
this is the time period when the network performance is experienced by the network users. The corresponding tools must be robust and
versatile to help in change the bandwidth of the link dynamically. In this project the student will apply methods of statistics and
probability theory to the available recorded real internet traffic data with the aim to evaluate the required bandwidth to provision a link.
The project will involve both theory and computational work, so some basic knowledge of probability theory and Matlab software is
required. The recorder network traces of the real traffic data from University Twente (Netherlands) will be used in this study.

Physics-Based Modelling Development and deployment of mobile sensor networks for intrusion detection is one of the hot technological topics stimulated by Available - can MSc Telecommunications
of the Intrusion Detection demands of security and various types of surveillance. Mobile sensors are capable to overcome limitations of stationary wireless sensor be carried out & Information Systems
in Mobile Sensor networks covering wider areas and providing mutual backup in case of failures. Due to random movements of mobile sensors and by 2 students
Networks unpredictable intruder trajectories the intruder detection process presents a classical problem closely related to physical stochastic
processes, such as, for example, collisions of the molecules in kinetic gas theory or mutual particle diffusion in mixtures. In this project the
student will develop a mobile wireless sensor network model using various random deployment strategies with different distributions of
sensors which are characterised a set of specific detection parameters. The resulting system of mobile sensors and intruders will then be
investigated both analytically and numerically with the goal to determine its optimal performance in the intruder detection subject to
various constrain conditions and comparing operation of mobile versus stationary sensor networks. The project will suit a student with
strong interest in novel sensor networks and application of physical stochastic approaches to mobile network analysis.

Development and The research project will involve development of theoretical model and simulation of wavelength tunable semiconductor lasers for Available MSc Telecommunications
Modelling of Wavelength telecommunication applications. The modelling will be carried using one of the software packages available in our group (either & Information Systems
Tunable Optical Crosslight, Synopsys, or VPI), both of which are capable to simulate these multi-section lasers. Good knowledge of optoelectronics is
Transmitters for required for successful work on the project together with basic knowledge of semiconductor device physics. No programming skills is
Telecommunication required. It is possible that the student will closely interact with one of our PhD students working at present in this area.
Applications

TCAD-Based Comparative The research project will involve development of theoretical model and numerical simulation of single wavelength semiconductor lasers Available MSc Telecommunications
Studies of Single Mode for telecommunication applications exploiting Technology Computer Aided Design (TCAD) approaches. The modelling will be carried using & Information Systems
Optical Transmitters for one of the software packages available in our group (Crosslight or Synopsys), both of which are capable to simulate advanced optical
Telecommunication devices. Good knowledge of optoelectronics is required for successful work on the project together with basic knowledge of
Applications semiconductor device physics. No programming skills are required. It is possible that the student will closely interact with one of our PhD
students working at present in this area.

Zhai Xiaojun A heterogonous The aim of this project is to design and implement a heterogonous computer vision system using a Zynq SoC. During this project, the Available All degrees
computer vision system student will need to first understand the architecture of the modern computer vision acquisition system, and then design and implement
using Zynq SoC the architecture in a heterogonous Zynq SoC using Vivado, Vivado HLS and/or SDSoC from Xilinx, e.g. Vivado HLS or SDSoC. At the end of
the project, the student needs to provide a number of IP cores that could be used to build a heterogonous computer vision in a Zynq SoC,
which may include one or few image processing filter(s) that could be used to verify the architecture. The student will need to understand
some fundamental computer vision architecture and algorithm as well as hardware/software codesign techniques. A strong C/C++
programming skills and computer architecture knowledge are essential to this project
Zynq SoC based deep The aim of this project is to design and implement a deep learning based object detection system on FPGA. During this project, the Available All degrees
learning network student will need to first understand the one of the existing deep learning frameworks and design the frame work in a PC, the target
implementation application could be a simple image processing or object detection problem. Once the network is design, it requires to move the deep
learning neural network to FPGA using one of the high-level synthesis tools from Xilinx, e.g. Vivado HLS or SDSoC. Finally, the student need
to verify the FPGA implementation results with the PC’s. By doing this project, the student will need to understand the principle of deep
learning neural network as well as hardware/software codesign techniques. A strong C/C++ programming skills and computer
architecture knowledge are essential to this project. The student will need to understand the principle of deep learning neural network as
well as hardware/software codesign techniques. A strong C/C++ programming skills and computer architecture knowledge are essential
to this project

Human Pose Estimation The aim of this project is to design and implement a deep learning based human pose estimation system. The project expects the student Available All degrees
with Deep Learning involves significant software developments, as well as implement the algorithm on an embedded GPU platform. Human Pose Estimation
is defined as the problem of localization of human joints (also known as key points - elbows, wrists, etc) in images or videos. It is also
defined as the search for a specific pose in space of all articulated poses. The student will need the following experience in order to
undertake this project. Python, or C++ programming skills are essential to this project, some CUDA programming experience is also
desirable. The student will also need to understand the principle of deep learning neural network.
Related readings: https://nanonets.com/blog/human-pose-estimation-2d-guide/
DeepPose: https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/42237.pdf

Energy-aware machine The aim of this project is to study, design and implement an energy-aware machine learning pruning algorithm. The current deep Available All degrees
learning pruning learning algorithm has been approved with significant redundancy, and it is not well suitable for embedded devices. In this Project, the
algorithm student will be focused on design an energy-aware machine learning pruning algorithm to remove redundancy of a typical deep learning
network, the algorithm includes removing the unnecessary weights, optimise the internal architecture of the network, approximating the
arithmetic, etc., the final optimised network will be operated at similar accuracy as the original neural network after pruning. The
student will need the following experience in order to undertake this project, good understanding of deep learning neural network, and
java, Python, C++ programming skills are essential to this project.
Related readings: Designing Energy-Efficient Convolutional Neural Networks Using Energy-Aware Pruning
http://openaccess.thecvf.com/content_cvpr_2017/html/Yang_Designing_Energy-Efficient_Convolutional_CVPR_2017_paper.html
A Survey of Model Compression and Acceleration
for Deep Neural Networks https://arxiv.org/pdf/1710.09282.pdf

Chip remote The aim of this project is to develop practical new techniques to enable monitoring of embedded hardware/software systems of varying Available All degrees
measurement and granularity and functional implementations. In this project, a new measurement tools will be developped, this will include dedicated
systems assessment circuits are designed in FPGA to monitor the hardware of systems and their interactions with extreme environments. Typically, this
circuits or debug infrastructures will operate in parallel with the system circuits and they will not normally impact the system's behaviour,
enabling non-intrusive measurement. The captured data at a lower level of operation will be used for behaviour analysis offline to
determine the behaviour of the system. The developed circuits will: operate autonomously without using the system’s processing
resources; enable analysis at runtime; provide access not only to the operation of the embedded hardware and software, but also the
processed sensor data. A strong Python, or C++ programming skills, embedded systems, FPGA, VHDL skills are required to undertake this
project. A strong Python, or C++ programming skills, embedded systems, FPGA, VHDL skills are required to undertake this project

Online device metrics The aim of this project is to develop practical new techniques to enable online operational assessment of device metrics derived directly Available All degrees
classifier from properties of the embedded hardware/software systems (ICMetrics). These metrics help identify any failures in the system, including
the hostile environment. This will be based on empirical techniques that use non-intrusive measurement points placed in the system, to
compare the actual behavioural characteristics, observed from software and hardware, against predicted characteristics. Systems will be
prototyped in a FPGA, with implementation of a real-time machine learning classifier to detect and assess the behaviour of the system. A
strong Python, or C++ programming skills, embedded systems, FPGA, VHDL skills are required to undertake this project.

END END END END

You might also like