Hyper-parameter Optimisation of Gaussian Process Reinforcement Learning for Statistical Dialogue Management.
Lu ChenPei-Hao SuMilica GasicPublished in: SIGDIAL Conference (2015)
Keyphrases
- hyperparameters
- gaussian process
- dialogue management
- model selection
- reinforcement learning
- cross validation
- bayesian framework
- bayesian inference
- closed form
- support vector
- random sampling
- em algorithm
- prior information
- noise level
- dialogue system
- sample size
- maximum likelihood
- maximum a posteriori
- learning agent
- incremental learning
- approximate inference
- incomplete data
- parameter settings
- missing values
- state space
- parameter estimation
- partially observable markov decision process
- feature selection
- expectation maximization
- parameter space
- markov decision processes
- learning algorithm
- bayesian networks
- parameter values
- machine learning
- computer vision
- regression model
- partially observable
- generative model
- worst case