Feb. 12, 2024, 5:43 a.m. | Jo\~ao A. C\^andido Ramos Lionel Blond\'e Naoya Takeishi Alexandros Kalousis

cs.LG updates on arXiv.org arxiv.org

In this paper, we introduce MAAD, a novel, sample-efficient on-policy algorithm for Imitation Learning from Observations. MAAD utilizes a surrogate reward signal, which can be derived from various sources such as adversarial games, trajectory matching objectives, or optimal transport criteria. To compensate for the non-availability of expert actions, we rely on an inverse dynamics model that infers plausible actions distribution given the expert's state-state transitions; we regularize the imitator's policy by aligning it to the inferred action distribution. MAAD leads …

adversarial algorithm availability cs.lg distribution expert games imitation learning novel paper policy sample signal trajectory transport

Research Scholar (Technical Research)

@ Centre for the Governance of AI | Hybrid; Oxford, UK

HPC Engineer (x/f/m) - DACH

@ Meshcapade GmbH | Remote, Germany

Senior Analyst-Data Analysis

@ Tesco Bengaluru | Bengaluru, India

Data Engineer - Senior Associate

@ PwC | Brussels

People Data Analyst

@ Version 1 | London, United Kingdom

Senior Data Scientist

@ Palta | Simple Cyprus or remote