Computationally Efficient PAC RL in POMDPs with Latent Determinism and Conditional Embeddings.
Masatoshi UeharaAyush SekhariJason D. LeeNathan KallusWen SunPublished in: ICML (2023)
Keyphrases
- computationally efficient
- reinforcement learning
- partially observable markov decision processes
- markov decision processes
- optimal policy
- continuous state
- state space
- policy gradient
- function approximation
- sample complexity
- latent space
- upper bound
- partially observable
- learning algorithm
- low dimensional
- multi agent
- reinforcement learning algorithms
- belief state
- euclidean space
- dynamic programming
- latent variables
- manifold learning
- learning classifier systems
- policy search
- learning problems
- vector space
- finite state
- pac learning
- sample size
- dimensionality reduction
- supervised learning
- machine learning
- computational complexity
- markov decision problems
- vc dimension
- belief space
- control policies
- rl algorithms
- model free
- actor critic
- distributed constraint optimization