You are on page 1of 57

Overview of Machine Learning

Adane Letta Mamuye (PhD)

June 2020
Outline

• Course Syllabus

• Overview of machine learning


Course Description

• Help you understand how state of the art machine learning


algorithms work.

• Design a machine learning approach to solve specific


problems.

• Understand the current research issues and trends in


Machine learning
Learning Objectives

• Upon successful completion of this course, students will:


– Abel to analyze and solve machine learning problems

– Compare different ML algorithms according to the


properties of their inputs and outputs.

– Get a research skills to locate and utilize the literature in


ML .
Topics to be Covered

• Machine learning overviews


• Supervised learning
• Unsupervised learning
• Deep learning
Evaluation

• Self and group evaluation


– Literature review – 20%
– Individual – Project I and group project (a team of two members) – Project II—
40%
– Final exam – 40%
• Assignment and Exams
– Literature review – July 06, 2020
– Project I- Implementing preprocessing algorithms – dimensionality reduction–
July 16, 2020
– Project II- Implementing machine learning algorithms --- July 30,2020
– Final Exam- ______
Machine Learning an Overview
“Any sufficiently advanced technology is
indistinguishable from magic.” – Arthur C.
Clarke (1961)
Learning

• Definition of learning:

– To get knowledge of by study, experience, or being taught;


– To become aware by information or from observation;
– To commit to memory;
– To be informed of ascertain;
– To receive instruction.

• These meanings have some shortcomings when it comes to


talking about computers.
Learning Problems

We wonder whether computers might be made to learn.

• How a computer learn to drive an autonomous vehicle?

Image src: autonomou

• How a computer learn to recognize handwritten words.


Learning Problem

• Imagine computers
– Learning from medical records which treatments are most
effective for new diseases

– Houses learning from experience to optimize energy costs


based on the particular usage patterns of their occupants

– Personal software assistants learning the evolving interests


of their users in order to highlight especially relevant
stories from the online morning newspaper.
What is learning

• Suppose we would like to program a machine that learns how


to filter spam e-mails.

– Memorize all previous e-mails that had been labeled as


spam e-mails by the human user
– When a new e-mail arrives-search for it in the set of
previous spam e-mails

• Is it a machine learning?
What is learning?
• It lacks an important aspect of learning systems the ability to
label unseen e-mail messages.

• A successful learner should be able to progress from


individual examples to broader generalization-- inductive
reasoning or inductive inference

• The incorporation of prior knowledge, biasing the learning


process, is inevitable for the success of learning algorithms

• The stronger the prior knowledge, the easier it is to learn


from further examples
The Science Witchcraft– a learning machine
ML-History

• ML grew out of the ambitious goal of artificial intelligence.

Image source: weblizar.com


ML-History

• In AI problems approached with various Symbolic methods,


neural networks, the generalized linear models of statistics
and probabilistic reasoning.

• Probabilistic reasoning faced a practical problem in data


acquisition.

• Expert systems had come to dominate AI, and statistics was


out of favor.
ML-History
• Machine learning, reorganized as a separate field, started to
flourish in the 1990s.

• The field changed its goal from achieving artificial intelligence


to tackling solvable problems of a practical nature.

• We do not yet know how to make computers learn nearly as


well as people learn.

• ML algorithms have been invented that are effective for


certain types of learning tasks.
Definition

• Definition

– In 1959, Arthur Samuel defined ML as a “Field of study that


gives computers the ability to learn without being explicitly
programmed”.

– Tom Mitchell (1998): “a computer program is said to learn


from experience E with respect to some class of tasks T and
performance measure P, if its performance at tasks in T, as
measured by P, improves with experience E.”
Machine Learning

• Goal:-

– Build a model from inputs in order to make data-driven


predictions or decisions, rather than following strictly static
program instructions.

- With machine learning we can gain insight from a dataset; we’re


going to ask the computer to make some sense from data.
Machine Learning

Taken from Eric: Intro. to


machine learning
When Do We Need Machine Learning?

• Two aspects of a given problem may call for the use of


programs that learn and improve on the basis of their
experience:

– Tasks That Are Too Complex to Program– humans do


routinely or tasks beyond human capability

– Adaptivity:- many tasks change over time or from one user to


another
Designing a learning system

• In order to complete the design of the learning system, we


must choose:

– The exact type of knowledge to be learned


Data
– A representation for this target knowledge
– A learning mechanism
Data – Key Ingredient of ML

• The component of experience, or training, in machine


learning often refers to data.

• In ML data is viewed as list of examples

• Collected from nature or industrial process


– Sorted in many forms: structured and unstructured
Data Everywhere

• Google: search 3.5 billion searches per day.

• Facebook: More than 250 billion photos have been uploaded


to Facebook. This equates to 350 million photos per day. -
2019
Data Everywhere

• Youtube: 300 hours of video are uploaded to YouTube every


minute! Almost 5 billion videos are watched on Youtube
every single day- Feb 2020.

• Twitter: Every second, on average, around 6,000 tweets are


tweeted on Twitter, which corresponds to over 350,000
tweets sent per minute, 500 million tweets per day and
around 200 billion tweets per year – 2019.
Data Types

• Data comes in different sizes and also types:


– Texts
– Numbers
– Click streams
– Graphs
– Tables
– Images
– Transactions
– Videos
– Some or all of the above!
Data Types from ML Perspective

• There are 4 basic types of data:


– Numerical: measurable can be continuous (height, weight,
temperature) or discrete (unit sold, no of language spoken).

– Categorical: can be nominal or ordinal (order matters)

– Time series: numbers collected at regular intervals over some


period of time

– Text: a collection of words


Machine Learning Paradigms

• Supervised versus Unsupervised:


– Learning tasks divided according to the nature of that
interaction with a teacher
– Spam e-mail detection
• Supervised learning describes a scenario in which the
experience contains significant information
– A teacher that supervises the learner by providing the extra
information (labels)
Machine Learning Paradigms

• Active vs passive learners


– An active learner interacts with the environment at training
time by posing queries
• Eg. Spam filter
– In an active setting, one could imagine asking users to label
spam e-mails chosen by the learner
Machine Learning Paradigms

• Online versus Batch Learning Protocol


– In many data mining settings, the learner the data miner, has
large amounts of training data to play with before having to
output conclusions
Kinds of ML

• ML tasks are typically classified into three broad categories


Supervised Learning

• The computer is presented with example inputs and their


desired outputs to learn a general rule that maps inputs to
outputs.
• The existence of a “teacher” who knows the right answer
– Goal: to find a good function f which maps (x1, x2, … xd) → y
by minimizing the error (loss) function.
– What is this function? How do we find it?
– What ever this function is, we would like to map out inputs
to the right output.
– Need some metric to tell us close we are to the ideal function

Supervised because we’re telling the algorithm what to predict


Supervised Learning

• Algorithms: Classification (Support Vector Machines, neural


networks, decision trees, K-nearest neighbors, naïve Bayes) ,
Regression

Image src: Supervise learning method


Supervised Learning

• Supervised learning problems are categorized by the type of


the output domain:
– If y = R, this problem is called regression

– If y is a categorical variable (part of a finite discrete set)–


called classification

– In general, y could be a lot more complex (graph, tree, …)


called structured prediction
Unsupervised learning

• No labels are given to the learning algorithm.


• Find interesting information or pattern in the input data.
• Example: clustering, dimensionality reduction, association
rule mining

Image src: unsupervised learning


Reinforcement learning
• A computer program interacts with a dynamic environment in
which it must perform a certain action.
– Learning what actions to take based on a reward signal

– Involves an assumption of being in a state S1 and evaluating the


chance of changing to another S2

– In the end (along the way) a reward signal is received.

– Want to take action to get the most reward possible

– The objective is to maximize the long-term cumulative reward.


Reinforcement learning
• The algorithm can collect data interactively

Image src: UT Computer Science

Try the strategy and Data Improve the strategy


collect feedbacks collection Training based on the
feedbacks
Issues in Machine Learning

• The field of machine learning is concerned with answering


questions such as:
– What algorithms exist for learning from specific training
examples?

– In what settings will particular algorithms converge to the


desired function, given sufficient training data?

– Which algorithms perform best for which types of problems


and data types?
Issues in Machine Learning

• How much training data is sufficient?

• How can the learner automatically alter its representation to


improve its ability to represent and learn the target function?
Application of ML

• Bioinformatics, Brain-machine interfaces, Classifying DNA


sequences, Computer vision, Object recognition

• Detecting credit card fraud, Information retrieval, Internet


fraud detection, Medical diagnosis, Natural language
processing, Recommender systems

• Search engines, Sentiment analysis (or opinion mining),


Sequence mining, Speech and handwriting recognition, Stock
market analysis, Weather forecast, spam filtering, MRI image
analysis, Handwritten recognition, Scene classification.
IoT and ML
• In IoT almost every electronic device are connected to a local
network or the internet.

Image src: Towards Data


Science

• A study conducted by Cisco estimates that IoT devices will generate


600 zetta bytes of data per year by 2020.
ML in Search Engines

Image src: Search Engine Journal


ML in Biomedicine

Image src: ASU Bioinformatic Core Lab


ML in Signal Analysis
ML in Image Recognition

• Face recognition problem:


Whose face is this in the
picture?
• Humans recognize with
experience: The more we see
the faster we perceive.

• Hard to model describing face


and its components
ML learning in Image processing

Image src: Katya


ML learning in Speech Recognition
ML Algorithms in Nutshell

• Tens of thousands of machine learning


algorithms
- Hundreds new every year
• Every ML algorithm has three components:
– Representation
– Optimization
– Evaluation
ML Algorithms in Nutshell
• Numerical functions
– Linear regression
– Neural networks
– Support vector machines
• Symbolic functions
– Decision trees
– Rules in propositional logic
– Rules in first-order predicate logic
• Instance-based functions
– Nearest-neighbor
– Case-based
ML Algorithms in Nutshell
• Probabilistic Graphical Models
– Naïve Bayes
– Bayesian networks
– Hidden-Markov Models (HMMs)
– Probabilistic Context Free Grammars (PCFGs)
– Markov networks
Various Search/Optimization
Algorithms
• Gradient descent
– Perceptron
– Backpropagation
• Dynamic Programming
– HMM Learning
– PCFG Learning
Various Search/Optimization
Algorithms
• Divide and Conquer
– Decision tree induction
– Rule learning
• Evolutionary Computation
– Genetic Algorithms (GAs)
– Genetic Programming (GP)
– Neuro-evolution
How to choose the right algorithm

• How can you choose which one to use?


– You need to consider your goal
• What are you trying to get out of this?
– Do you want a probability that it might rain tomorrow, or
do you want to find groups of voters with similar
interests?
• What data do you have or can you collect?
– If you’re trying to predict or forecast a target value, then you
need to look into supervised learning.
– If not, then unsupervised learning is the place you want to
be.
How to choose the right algorithm

• How can you choose which one to use?


– You should spend some time getting to know your data, and
the more you know about it, the better you’ll be able to build
a successful application.
– With the algorithm narrowed, there’s no single answer to
what the best algorithm is or what will give you the best
results.
• You are going to have to try different algorithms and
see how they perform.
Maths for Machine Learning

• Equations, Functions, and Graphs

• Differentiation and Optimization

• Linear algebra- vectors and Matrices

• Statistics and Probability


– Measures of central tendency, Spread of the data,
Distributions, Hypothesis testing, etc.
References

• Neural Networks and Deep Learning by Michael Nilson


• The Elements of Statistical Learning by Trevor Hastie, Robert
Tibshirani & Jerome Friedman
• Machine Learning: A Probabilistic Perspective Kevin P.
Murphy, MIT Press, 2012.
• Pattern Recognition and Machine Learning Christopher M.
Bishop, Springer, 2006.
• Ian H. Witten and Eibe Frank . Data Mining: Practical Machine
Learning Tools and Techniques, 2005.
• Josh Patterson and Adam Gibson. Deep Learning: A
Practitioner’s Approach, 2017.
Thank you

You might also like