Skip to main content

Using behavioural data to model a game-theoretic adversary


Suitable for

MSc in Computer Science


To take the project, a student should have done the courses in Computational Learning Theory, and Machine Learning.

The project is related to the paper (Yang et al.) below, and uses a data set (studied in the paper) obtained from an on-line game that simulates an adversarial security setting, in which the player selects one of a number of targets to attack, based on information about their values, and the probabilities that they will be defended. The purpose of this game is to understand a player's "bounded rationality", i.e. the ways in which he fails to make the best decision. The project would study a model of player behaviour, related to ones studied in (Yang et al.), and aim to learn the ways in which a player fails to best-respond. A related objective is to design optimal defender behaviour based on what is learned about attackers' behaviour.

@article{YKOTJ13, author="R.\ Yang and C.\ Kiekintveld and F.\ Ord{\'o}{\~n}ez and M.\ Tambe and R.\ John", title="Improving resource allocation strategies against human adversaries in security games: An extended study", journal="Artificial Intelligence", volume="195", pages="440--469", year="2013", }