Theme: Machine Learning
Course Title: A Decision Making View of Machine Learning

Lecturer: Dr Hanna Kurniawati

Course Content
This course will cover recent advances in decision making under uncertainty and its application to machine learning, in particular reinforcement learning. We will provide an overview of current work on Markov Decision Processes (MDPs), Partially Observable Markov Decision Processes (POMDPs), and Bayesian Reinforcement Learning. We will focus on both basic concepts and state of the art algorithms and computational techniques that made these approaches become practical.This course will also include hands-on demo using available software tools.

Background Reading

  • Ghavamzadeh, M., Mannor, S., Pineau, J., & Tamar, A. (2015). Bayesian reinforcement learning: A survey. Foundations and Trends® in Machine Learning,8(5-6), 359-483
  • Kurniawati, H., Hsu, D., & Lee, W. S. (2008, June). SARSOP: Efficient Point-Based POMDP Planning byApproximating Optimally Reachable Belief Spaces. InRobotics: Science and systems (Vol. 2008)
  • Silver, D., & Veness, J. (2010). Monte-Carloplanning in large POMDPs. InAdvances in neuralinformation processing systems (pp. 2164-2172)
Course Title: Martingales, McDiarmid and Machine Learning: How to validate models like a pro!

Lecturer: Dr Brendan van Rooyen

Course Content
Machine Learning can be broadly understood as the science of prediction and has applications right across the spectrum, from fraud detection to drug discovery, from natural language translation to understanding the human brain! Models of ever increasing complexity are now being “trained” on data sets of ever increasing size. Even with this explosion in scale, the age-old problem of model validation remains. Standard rules of thumb include cross validation and the bootstrap. While these are fantastic general purpose tools, they are computationally expensive and can be overkill for some problems.

In this course, we will explore the underlying mathematics of model validation, where we will lay bare the underlying assumptions underpinning learning. Come for an introduction to concentration inequalities, martingales and to find out what Colin McDiarmid has to do with Machine Learning.

Not limited to Machine Learning model validation, concentration inequalities have broader application across computer science and mathematics, where they facilitate a powerful set of tools for creating efficient algorithms.

Background Reading

  • None