Reinforcement learning with limited reinforcement: using Bayes risk for active learning in POMDPs.
Finale DoshiJoelle PineauNicholas RoyPublished in: ICML (2008)
Keyphrases
- reinforcement learning
- bayes risk
- active learning
- learning algorithm
- upper and lower bounds
- transfer learning
- learning process
- machine learning
- loss function
- supervised learning
- partially observable markov decision processes
- continuous state
- optimal policy
- reproducing kernel hilbert space
- markov decision processes
- state space
- distortion measure
- real valued
- posterior probability
- dynamic programming
- random sampling
- labeled data
- semi supervised
- training examples
- unlabeled data
- semi supervised learning