Survey of Multiagent Reinforcement Learning
Highlights of a survey paper on multiagent RL from 2008
Blog
Recent posts
Transition-Independent Decentralized MDPs
The paper is available here: Becker et al. 2004
Stabilising Experience Replay for MARL
Making experience replay work in non-stationary multi-agent settings
Deep Recurrent Q-Networks (DRQN)
Deep Q-Learning with an LSTM, for partially observable MDPs
Value Iteration Networks (VINs)
A fully differentiable planning module which can learn to plan end to end using backpropagation - NIPS 2016 Best Paper
Deep Deterministic Policy Gradients
A model-free actor-critic algorithm for continuous control that incorporates experience replay and target networks from DQN to the actor-critic approach to p...
Opponent Modeling in Deep Reinforcement Learning
Learning policies which adapt to the opponent’s strategy by giving information about the opponent along with the state as input to a Deep Q network