Benchmarking Uncertainty Estimates with Deep Reinforcement Learning for Dialogue Policy Optimisation.
Christopher TeghoPawel BudzianowskiMilica GasicPublished in: ICASSP (2018)
Keyphrases
- reinforcement learning
- optimal policy
- sequential decision making
- policy search
- markov decision process
- action selection
- partial observability
- reinforcement learning problems
- reinforcement learning algorithms
- actor critic
- partially observable
- markov decision processes
- control policy
- state space
- control policies
- policy gradient
- function approximation
- markov decision problems
- function approximators
- action space
- temporal difference
- policy iteration
- state and action spaces
- confidence intervals
- reward function
- man machine
- dynamic programming
- infinite horizon
- partially observable markov decision processes
- model free
- continuous state spaces
- approximate dynamic programming
- continuous state
- state action
- average reward
- state dependent
- dialogue system
- uncertain data
- partially observable environments
- learning algorithm
- dialogue management
- agent receives
- mixed initiative
- measurement error
- spoken dialogue systems
- natural language
- multi agent
- genetic algorithm
- policy evaluation
- machine learning
- finite state
- belief functions
- optimal control
- inverse reinforcement learning
- decision problems