Applied Machine Learning

About

Course Overview

The Michaelmas term of Social Data Science introduces students to the mathematical and statistical foundations of Machine Learning. This course extends this learning with practical and applied Machine Learning techniques. This will extend the mathematical foundations towards domains where we are uncertain about the right answer or best approach.

This course emphasizes applications in a variety of domains and focuses foremost on strategies for high quality results from data.

Topics introduced include:

• Creation of training and test datasets with techniques such as weak supervision;
• Transfer learning and fine-tuning models; best practices for annotation;
• Bayesian approaches to estimation;
• Optimisation of hyperparameters using Bayesian optimisation; and
• Techniques for interpretability and explainability.

Applications of the techniques covered will be bespoke and draw upon the skill set of the faculty, which may include applications to computer vision, speech, and language processing.

Learning Outcomes

• Understand machine learning algorithms for standard tasks (such as classification and regression) in such a way as to be confident in their results.
• Critically assess these methods (e.g., their internal workings versus claims made about their performance)
• Understand base assumptions of applications of machine learning algorithms versus their use in practice
• Understand how to engage with the social context of methods and applications