The University of Southampton
Courses

# COMP3223 Foundations of Machine Learning

## Module Overview

Machine Learning is about extracting useful information from large and complex datasets. The subject is a rich mixture of concepts from function analysis, statistical modelling and computational techniques. The module will cover the fundamental principles in the subject, where you will learn the theoretical basis of how learning algorithms are derived and when they are optimally applied, and gain some hands-on experience in laboratory-bases sessions. It will lead to more advanced material covered in later modules. Exclusions: Cannot be taken with COMP3206 or COMP3222 or COMP6229 or COMP6245 or COMP6246.

### Aims and Objectives

#### Learning Outcomes

##### Knowledge and Understanding

Having successfully completed this module, you will be able to demonstrate knowledge and understanding of:

• Underlying mathematical principles from probability, linear algebra and optimisation
• The relationship between machine learning and biological learning
##### Subject Specific Intellectual and Research Skills

Having successfully completed this module you will be able to:

• Characterise data in terms of explanatory models
• Use data to reinforce one/few among many competing explanatory hypotheses
• Gain a broad understanding of the latest research issues
##### Subject Specific Practical Skills

Having successfully completed this module you will be able to:

• Systematically work with data to learn new patterns or concepts
• Gain facility in working with algorithms to handle data sets in a scientific computing environment

### Syllabus

Historical Perspective - Biological motivations: the McCulloch and Pitts neuron, Hebbian learning. - Statistical motivations Theory - Generalisation: What is learning from data? - The power of machine learning methods: What is a learning algorithm? What can they do? Probability - Probability as representation of uncertainty in models and data - Bayes Theorem and its applications - Law of large numbers and the Multivariate Gaussian distribution Optimisation - Convexity - 1-D minimisation - Gradient methods in higher dimensions - Constrained optimisation Linear Algebra - Using matrices to find solutions of linear equations - Properties of matrices and vector spaces - Eigenvalues, eigenvectors and singular value decomposition Supervised Learning - Regression Analysis - Classification using Bayesian principles - Perceptron Learning - Support Vector Machines and introduction to Kernel methods - Neural networks/multi-layer perceptrons (MLP) - Features and discriminant analysis Data handling and unsupervised learning - Principal Components Analysis (PCA) - K-Means clustering - Spectral clustering Regression and Model-fitting Techniques - Linear regression - Polynomial Fitting - Kernel Based Networks Case Studies - Example applications: Speech, Vision, Natural Language, Bioinformatics.

### Learning and Teaching

#### Teaching and learning methods

Lectures, labs and guided self-study

TypeHours
Revision10
Supervised time in studio/workshop6
Follow-up work10
Preparation for scheduled sessions10
Lecture20
Total study time150

Bishop, Christopher M.. Pattern Recognition and Machine Learning.

Mackay, David J. C.. Information Theory, Inference and Learning Algorithms..

### Assessment

#### Summative

MethodPercentage contribution
Continuous Assessment 20%
Final Assessment  80%

#### Repeat

MethodPercentage contribution