Deep Generative Models for Imitation Learning and Fairness

Опубликовано 3 декабря 2018, 22:59

In the first part of the talk, I will introduce Multi-agent Generative Adversarial Imitation Learning, a new framework for multi-agent imitation learning for general Markov games, where we build upon a generalized notion of inverse reinforcement learning. Multi-agent settings are challenging due to the existence of multiple (Nash) equilibria and non-stationary environments. Our method can be used to imitate complex behaviors in high-dimensional environments with multiple cooperative or competing agents.

In the second part of the talk, I will discuss an information-theoretically motivated objective for learning maximally expressive representations subject to fairness constraints. This objective generalizes a range of existing approaches. We introduce a dual optimization method that allows the user to explicitly control the level of fairness. Empirical evidences suggest that our proposed method can account for multiple notions of fairness and achieves higher expressiveness at a lower computational cost.

View presentation slides and see more at microsoft.com/en-us/research/v...