Artificial and machine intelligence are increasingly reliant on models of the world that account for uncertainty - probabilistic models. When a robot moves around it is constantly trying to make inferences about the world based on prior beliefs, but also new data. The new data are unreliable, and the prior beliefs may be only approximate or weighted beliefs; the robot needs to be able to account for all the information it has to estimate the probability of success if it takes a certain course of action. Similarly, in medicine, we are constantly faced with problems of disease diagnosis (or trying to establish the cause of a disease), and we only have probabilistic information about the most likely disease (or the most likely cause). This module will look at all the tools and principles behind progress in these and other challenging problems.
Learning Outcomes
By the end of the module students should be able to:
Demonstrate a sound understanding of probability and its role in modelling, simulation, machine learning and robotics to include general models such as Gambler’s Ruin and drift
Select appropriate probabilistic models and formulate problems probabilistically in terms of random variables, sample space, events, conditional probability, expectations, bounds, and probability distributions
Apply randomized methods for decision making such as Monte Carlo, MCMC, and Monte Carlo Tree Search
Model and analyse complex data using probabilistic graphical models such as Bayesian networks, causal Bayesian networks, and approximate Bayes net inference