MSc ACS: Digital Biology / Course details

The world is filling up with data. Machine Learning is concerned with building mathematical models from this data, capable of tasks that would normally require a human. Typical applications might be spam filtering, speech recognition, medical diagnosis, or weather prediction. The data structures we use are known as "models" come in various forms, e.g. trees, graphs, algebraic equations, and probability distributions. The emphasis is on constructing these models automatically from data---for example making a weather predictor from a datafile of historical weather patterns. This course unit will introduce you to the concepts behind various Machine Learning techniques, including how they work, and use existing software packages to illustrate how they are used on data.

• To introduce the main algorithms used in modern machine learning.
• To introduce the theoretical foundations of machine learning.
• To provide practical experience of applying machine learning techniques.

If you have sat an undergraduate ML course (particularly my COMP24111) then you may feel you know all this material. In fact we will cover virtually the same topics - however, you almost certainly will not have covered this material in the same depth as we will cover it. We will study why and how these methods work, at a very deep level. This is not a course on how to use ML techniques. It is a course on the foundations, the deeper aspects. If you really think you know it all already, then try sitting the previous exam papers, under exam conditions of course (i.e. no textbooks).

• Have knowledge and understanding of the principle algorithms used in modern machine learning, as outlined in the syllabus.

• Have sufficient knowledge of information theory and probability theory to understand some basic theoretical results in machine learning.

• Be able to apply machine learning algorithm to real datasets, evaluate their performance and appreciate the practical issues involved.

• Be able to provide a clear and concise description and justification for the employed experimental procedures.

Topics covered:

• Classifiers and the Nearest Neighbour Rule
• Linear Models, Support Vector Machines
• Algorithm assessment - overfitting, generalisation, comparing two algorithms
• Decision Trees, Feature Selection, Mutual Information
• Probabilistic Classifiers and Bayes Theorem
• Combining Models - ensemble methods, mixtures of experts, boosting
• Feature Selection - basic methods, plus some tasters of research material

Lectures

1 day per week (5 weeks)

• Formative exam practice questions with answers and discussion
• Virtual face-to-face feedback on lab work