Learning from Data - 2019 entry

Artificially intelligent machines and software must assimilate data from their environment and make decisions based upon it.  Likewise, we live in a data-rich society and must be able to make sense of complex datasets.  This module will introduce you to machine learning methods for learning from data. You will learn about the principal learning paradigms from a theoretical point of view and gain practical experience through a series of workshops. Throughout the module, there will be an emphasis on dealing with real data, and you will use, modify and write software to implement learning algorithms. It is often useful to be able to visualise data and you will gain experience of methods of reducing the dimension of large datasets to facilitate visualisation and understanding.

Prerequisite module: ECM1400, ECM1415 or ECM1701

This module aims to equip you with the fundamentals of machine learning in a computer science context. It will provide a thorough grounding in the theory and application of machine learning and statistical techniques for classification, regression and unsupervised methods.  We will pay particular attention to methods for visualising complex datasets.

On successful completion of this module, you should be able to:

Module Specific Skills and Knowledge:

1 apply principles for statistical pattern recognition to novel data;
2 analyse novel pattern recognition and classification problems, establish models for them and write software to solve them;
3 utilise a range of supervised and unsupervised pattern recognition and machine learning techniques to solve a wide range of problems.
 

Discipline Specific Skills and Knowledge:

4 state the importance and difficulty of establishing principled models for pattern recognition;
5 use Matlab or other programming languages for scientific analysis and simulation.
 

Personal and Key Transferable / Employment Skills and Knowledge:

6 identify the compromises and trade-offs that must be made when translating theory into practice;
7 critically read and report on research papers.

- introductory material: Basic ideas of classification, regression and preliminary concepts;

- data description;

- statistical preliminaries: probability, standard distributions and densities, Bayes rule;

- modelling and learning: models, noise, maximum likelihood learning and error functions; generalisation; common neural network architectures;

- optimisation for learning;

- classification: decision boundaries; k-nn classifier, linear discriminants, kernel-based classifiers, kernel trick, large margin classifiers;

- receiver operating characteristics: loss functions; ROC curves and their optimisation;

- unsupervised methods: clustering, k-means;

- dimension reduction and visualisation: PCA, ICA, linear and nonlinear methods for visualisation, MDS and isomap;

- feature extraction: sequential forwards/backwards selection;

- learning systems with temporal coupling: hidden Markov models; object tracking in video.

Two workshops will be assessed.  Other workshops will have formative components.

Referred and deferred assessment will normally be by examination. For referrals, only the examination will count, a mark of 40% being awarded if the examination is passed. For deferrals, candidates will be awarded the higher of the deferred examination mark or the deferred examination mark combined with the original coursework mark.

Reading list for this module: