Rewriting History with Inverse RL: Hindsight Inference for Policy Improvement.
Benjamin EysenbachXinyang GengSergey LevineRuslan SalakhutdinovPublished in: CoRR (2020)
Keyphrases
- action selection
- optimal policy
- reinforcement learning
- markov decision process
- control policy
- markov decision processes
- partially observable domains
- action space
- policy iteration
- significant improvement
- bayesian networks
- markov decision problems
- control policies
- state space
- probabilistic inference
- bayesian inference
- rewriting rules
- policy evaluation
- rewrite rules
- inference process
- conjunctive queries
- integrity constraints
- function approximators
- belief networks
- state action
- rl algorithms
- average cost
- function approximation
- policy gradient
- infinite horizon
- policy search
- state and action spaces