Learning to act in multiagent systems is distinctive from traditional single-agent learning, in that the optimal strategy depends on others: an agent seeks to find the best response strategy and the other agents may adapt their strategies in turn. Our goal is to tackle partially observable multiagent scenarios by proposing a framework based on learning robust best responses (i.e., skills) and Bayesian inference for opponent detection. In order to reduce long training periods, we propose to intelligently reuse policies (skills) by quickly identifying the opponent we are playing with.
Our NeurIPS 2021 Reading List
Computer Vision; Data Visualization; Graph Representation Learning; Learning And Generalization; Natural Language Processing; Optimization; Reinforcement Learning; Time series Modelling; Unsupervised Learning
Heterogeneous Multi-task Learning with Expert Diversity
Computer Vision; Natural Language Processing; Reinforcement Learning
Robust Risk-Sensitive Reinforcement Learning Agents for Trading Markets