Giorgia Ramponi

Active policy selection, not just expert samples, is essential for strategically identifying rewards in multi-agent games, enabling near-optimal downstream decision-making.

Forget state-action spaces: this work achieves efficient multi-agent imitation learning by concentrating on feature-level representations in linear Markov games.

Stop hand-tuning reward learning losses: MAVRL learns a shared reward function from diverse feedback signals by treating each as a likelihood within a Bayesian inference framework.