Rock-paper-scissors for the computerised strategist
In the well-known game of rock-paper-scissors, it is clear that any player can "break even" by playing entirely at random. On the other hand, people do a poor job of generating random numbers, and expert players of the game can take advantage of predictable aspects of opponents' behaviour. In this project, we envisage designing algorithms that adapt to human opponent's behaviour, using for example no-regret learning techniques, and modelling the opponent as a probabilistic automaton. Ideally, the student taking this project should manage to persuade some volunteers to test the software! This is needed to provide data for the algorithms, and should provide feedback to the opponent on what mistakes they are making. Depending on how it goes, we could also consider extending this approach to related games.
Prerequisite: Interest in working with probability is important.