You are on page 1of 11

FYP selection

Profs to avoid: Loke Yuan Ren, Sourav Bhouvik, Owen Newton, Zinovi

Profs to go for: Chnh Eng Siong, Sourav Sengupta, Li Fang

Areas of focus: blockchain, financial modelling

Rank Title Prof name Description


1 Network Analysis and Dr Sourav Sen In this project, you will explore the existing
Traceability on Ethereum Gupta research developments on network analysis and
Blockchain traceability in the Ethereum blockchain, and design
an analytics toolbox for the same. The goal of the
project will be to implement an end-to-end
analytics toolbox to perform data mining, network
analysis and traceability analysis on Ethereum.

It is expected that you will have a basic


understanding of data mining, and will be a
competent programmer. You should be willing to
learn theoretical foundation behind network
analysis and traceability. You should also be able to
adapt open-source code or packages for the target
implementation. No need to know about
blockchain.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component

Study the existing literature on Ethereum network


mining and traceability in order to find an open
problem.

(b) Development component

Develop an end-to-end analytics toolbox to


perform data mining, network analysis and
traceability analysis.
2 Network Analysis and Dr Sourav Sen In this project, you will explore the existing
Traceability on Bitcoin Gupta research developments on network analysis
Blockchain and traceability in the Bitcoin blockchain, and
design an analytics toolbox for the same. The
goal of the project will be to implement an
end-to-end analytics toolbox to perform data
mining, network analysis and traceability
analysis on Bitcoin.

It is expected that you will have a basic


understanding of data mining, and will be a
competent programmer. You should be willing
to learn theoretical foundation behind network
analysis and traceability. You should also be
able to adapt open-source code or packages for
the target implementation. No need to know
about blockchain.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component

Study the existing literature on Bitcoin


network mining and traceability in order to
find an open problem.

(b) Development component

Develop an end-to-end analytics toolbox to


perform data mining, network analysis and
traceability analysis.
8 Quantitative Analysis of Smart Li Yi Smart contracts are software programs that
Contracts (Vulnerability) execute on top of blockchains to manage large
sums of money, carryout transactions of assets,
and govern the transfer of digital rights
between different parties. Despite the high
stakes involved in these activities, smart
contracts are often written in a very
undisciplined way thus far. Many high-profile
vulnerabilities associated with significant
financial implications have been reported
recently. The existence of vulnerabilities
compromise the security and reliability of
smart contracts, and greatly endanger the trust
of participating parties in their ongoing
businesses. The aim of this project is to study
different types of vulnerabilities in smart
contracts and understand their causes, and
eventually build automatic tools for
discovering vulnerabilities.

Student should be familiar with compiler


basics.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component

Understand Ethereum smart contracts, model


their behaviors, apply analysis tools/techniques
on the models.

(b) Development component

Develop program analysis tools for Ethereum


Virtual Machine (EVM) bytecode for
vulnerability detection.
5 Application of Machine Yeo Chai Kiat WWith the advance in data analytics and deep
Learning in the Forecast of learning technology, the aim of this project is to
Financial Markets investigate the application of artificial neural
networks (deep learning) for prediction. The
student shall identify the financial market which
he/she wishes to predict, eg. stock market index,
individual stocks etc. Student shall use multi-modal
data to improve the accuracy of the prediction.
Examples are a combination of technical
indicators, news reports, social media. Sentiment
Analysis is to be incorporated too.

Skills required:
a. Strong software and algorithmic skills
b. Machine learning techniques
c. AI techniques
d. Data Analytics techniques
10 Federated learning on Ast/P Liu Federated learning has been proven promising in
heterogeneous edge devices Weichen training deep neural networks with limited training
data, which employs multiple devices to train a
common deep neural network. In many cases, the
data owned by different clients are private and
cannot be shared with other clients, such as the
users’ financial information from different banks.
In federated learning, there is a server and
multiple clients, the server is responsible for
coordinating different clients. First, the server
offloads a deep neural network to all clients, then
each client starts to train the network with their
own data without sharing the data with other
clients. Finally, all clients upload the training
information, such as gradients, to the server and
the server integrates this information into the
model for higher performance.

However, federated learning on edge devices is


more complex since different clients possess
different computing resources (i.e. memory and
processors). It is still an open problem to modify
and adjust the model based on the hardware
constraints. This project will investigate the
feasibility of using existing model compression
techniques to modify the networks according to
the hardware, such that we can reduce the
resource consumption and boost the performance
of the training process.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component

1. Understand and research different techniques


on federated learning
2. Understand model compression techniques
(pruning, quantization, etc.)

(b) Development component

1. Implement different model compression


algorithms
2. (Optionally) Integrate compression algorithms
into federated learning
4 AI for Finance A/P Erik A major application of AI research is in the
Cambria
financial domain. In this FYP, we will use
opinion mining to translate natural language
data in numerical data that can be fused with
other structured data coming from the stock
market.

Over the last decades, the increasing capability


of NLP makes it possible to capture market
sentiments more accurately and the semantics
of financial documents in a more nuanced way,
hence nurturing the research cluster of natural
language based financial forecasting. The
influence of such alternative data is not just
limited to the prediction of stock and
commodity prices but also prevails portfolio
management, risk control, pricing, and other
investing activities.

To bridge the gap between finance theories and


the practice of text mining, this FYP
specifically looks at the distribution of asset
returns, the correlations between different
assets, and how to preserve their desired
properties such as robustness and stationarity
in high-dimensional cases. The FYP also
includes investigations in fields such as
volatility forecasting, asset allocation, and
market trend prediction.

Specific details:
Development of intelligent applications for
finance

Development of AI algorithms in the context


of financial research
3 AI for Business Intelligence A/P Erik One of the prime applications of sentiment
Cambria analysis is business intelligence. In this FYP, we will
apply AI for enhancing decision-making in contexts
such as brand positioning and social media
marketing.

The evolution of the Web from a read-only to a


read-write entity made users more enthusiastic
about interacting and collaborating through social
networks, blogs, wikis and other on-line
collaborative media. We will use AI to extract
sentiment information from such data. Not only
textual data but also visual data, by means of
image sentiment analysis, and social network data,
by means of community embeddings.

We will also explore how the ensemble application


of sentiment analysis and user modeling can be
exploited to create better recommendation
systems. In particular, we will use deep neural
networks to better learn both user and item
representations and make these close to binary
codes such that the quantization loss is minimized.
In addition, we will extend the proposed
framework for out-of-sample cases, i.e., dealing
with new users, new items, and new ratings.

Specific details:
Development of applications for business
intelligence

Development of AI algorithms in the context of


sentiment analysis research
9 AI based stock market Dr Li Fang Stock market trending analysis has been one of the
trending analysis most interest research area. Given the dynamic
nature of stock market and its relationship follows
non-linearly, various theories have highlighted the
unattractiveness of stock market prediction,
making it as a redundant task. With the advent of
AI technology, more and more researchers put
efforts on AI based financial market analysis
methods and financial market analysis becomes
possible. However, it is still a challenging task to
forecast stock market trending accurately.

Specific details:

Student Expectations
In this project, the student will be guided to learn
the existing stock market trending and predicting
methods and analyze the relationship between the
parameters and the performance of the trending
algorithms, with the aim to enhance the existing
methods to develop new methods for stock
market trending analysis.
The students have a background in data mining, or
machine learning, or have knowledge of
programming languages, Python is preferred.
This is a collaboration work with SMU. The student
with work with both NTU supervisor and a SMU
professor.
6 BlockFL: Blockchain-enabled Prof Dusit Recently, considering the increasingly stringent
decentralized Federated Niyato data privacy legislation and growing privacy
Learning and model trading concerns, the concept of Federated Learning (FL)
has been introduced. In Federated Learning,
clients can use their locally stored data to train an
AI model required by a server. The client can then
send the model updates, rather than raw data to
the server for aggregation. FL enables us to tap on
the wealth of data collected in smartphones,
smart-homes, and intelligent vehicles.

However, conventional Federated Learning


implementation requires a centralized entity to
conduct and coordinate the training process. This
is prone to a single point of failure, e.g., when the
centralized entity is malicious. In this project, a
blockchain based FL is proposed to facilitate
decentralized FL. The FL model training is carried
out on a blockchain. Then, the derived global
model is traded on a cross chain platform. The
BlockFL platform is also expected to be tested on
specific uses, e.g., healthcare and finance
applications.

Specific details:
(a) Design component

1) The platform enables the secure aggregation of


model parameters from multiple users

2) The platform supports model compression,


lightweight model training, and efficient
aggregation.

3) The cross-chain platform enables model trading.

(b) Implementation component

The prototype is expected to be delivered for


some simple testing, for example, in healthcare
and finance applications.

(a) Research component


(b) Development component
3 History Analysis of Li Yi Decentralized applications (DApps) or smart
Decentralized Applications contracts are computer programs running on
Blockchains. The historical transaction data of
all DApps is completely transparent and may
be used as a source of data for better
understanding of the DApp itself. In this
project, we build automated data analytics
techniques in conjunction to existing program
analysis tools to solve practical problems in
analyzing smart contracts.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component

Design algorithms to process real DApp


historical data and gain valuable insights.

(b) Development component

Implement automated tools for large-scale


analysis.
2 Opinion based Intelligent Dr Li Fang Recommender systems are technologies used
Recommender System to provide recommendations to users according
to their requirements. Traditional
recommender systems generally provide
recommendations based on users’ preference
history, but with the advent of social media
and AI technologies, it is possible to provide
richer information and recommendations by
making use of such social media information
and AI technologies.

Specific details:

Student Expectations
In this project, we will focus on developing an
intelligent opinion-based recommender
system. For the system to perform efficiently
and accurately, intelligent social media opinion
analysis must be performed on the acquired
data to process and classify the data into
different categories, such as positive, neutral
and negative. Based on the opinions obtained
from social media, the proposed system will be
able to recommend brands, products, services
or other useful guides or recommendations to
users.
The students have a background in data
mining, or machine learning, or have
knowledge of programming languages, Python
is preferred.
This is a collaboration work with SMU. The
student with work with both NTU supervisor
and a SMU professor.
6 Event detection from social A/P Sun Aixin This project is to research on event detection
media on Covid-19 algorithms from social media messages on covid-
19. The task is to detect the trending topics
discussed on social media along the development
of the covid-19.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component

(b) Development component


2 Design and development a A/P Sun Aixin This project aims to develop a unified demo
NLP demo system system to illustrate the functionalities and
differences from various NLP packages like NLTK,
spaCy, Standford NLP and others. The purpose is to
show the different implementations for various
NLP tasks like Tokenization, POS tagging, Parsing,
and others.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component


(b) Development component
3 An evaluation of A/P Sun Aixin This project requires the student to evaluate
tokenizers on domain various tokenizers offered in opensource toolkits
specific text like Lucene, NLTK on domain specific text.
Examples of domain specific text are text from
social media, text from specific topic domains like
medial, chemical texts.

Specific details:
(a) Design component

(b) Implementation component

(a) Research component

(b) Development component

SCSE21-0412 Network Analysis and Traceability on Ethereum Blockchain

SCSE21-0411 Network Analysis and Traceability on Bitcoin Blockchain

SCSE21-0231 AI for Finance

SCSE21-0166 Application of Machine Learning in the Forecast of Financial Markets

SCSE21-0078 Design and development a NLP demo system

SCSE21-0327 History Analysis of Decentralized Applications

SCSE21-0270 Opinion based Intelligent Recommender System

SCSE21-0077 An evaluation of tokenizers on domain specific text

SCSE21-0232 AI for Business Intelligence

SCSE21-0079 Event detection from social media on Covid-19

SCSE21-0198 BlockFL: Blockchain-enabled decentralized Federated Learning and model trading


SCSE21-0271 AI based stock market trending analysis
SCSE21-0172 Federated learning on heterogeneous edge devices

SCSE21-0123 Quantitative Analysis of Smart Contracts (Vulnerability)

You might also like