PhD Seminar • Artificial Intelligence | Autonomous Vehicles • Revealed Multiobjective Utility Aggregation Using Model Rationalizability

Friday, July 29, 2022 12:00 pm - 1:00 pm EDT (GMT -04:00)

Please note: This PhD seminar will take place online.

Atrisha Sarkar, PhD candidate
David R. Cheriton School of Computer Science

Supervisor: Professor Krzysztof Czarnecki

To create game theoretic models of human behaviour from observational data, a game designer needs to infer the utilities of the agents in the game. However, many human tasks are multiobjective. For example, during driving humans balance different potentially conflicting objectives, such as safety, progress, and comfort, in the process of selecting their desired action. The manner in which humans aggregate these objectives is often context dependent and individual specific. Developing a methodology for estimation of the parameters involved in the aggregation process is a necessary first step towards constructing the agent utilities in a game.

In this talk I will present different approaches and algorithms based on the ideas of rationalizability to estimate parameters involved in multiobjective aggregation that are specific to the underlying reasoning model and the aggregation method used by the agent.