Rethinking Action Spaces for Reinforcement Learning in End-to-end Dialog Agents with Latent Variable Models.
Tiancheng ZhaoKaige XieMaxine EskénaziPublished in: NAACL-HLT (1) (2019)
Keyphrases
- end to end
- action space
- latent variable models
- real valued
- reinforcement learning
- single agent
- action selection
- state space
- multi agent
- latent variables
- markov decision processes
- state and action spaces
- multiple agents
- continuous state
- multi agent systems
- learning problems
- latent dirichlet allocation
- function approximators
- decision problems
- probabilistic model
- dynamic programming
- topic models
- markov decision problems
- hidden markov models
- markov decision process
- stochastic processes
- machine learning
- path finding
- finite state
- hidden variables
- dynamic environments
- optimal policy
- computational complexity
- learning algorithm