Off-Dynamics Reinforcement Learning: Training for Transfer with Domain Classifiers.
Benjamin EysenbachSwapnil AsawaShreyas ChaudhariRuslan SalakhutdinovSergey LevinePublished in: CoRR (2020)
Keyphrases
- reinforcement learning
- transfer learning
- training set
- test set
- training examples
- domain specific
- cross domain
- training samples
- learning classifier systems
- dynamical systems
- machine learning algorithms
- supervised learning
- supervised training
- state space
- recurrent networks
- training data
- function approximation
- domain independent
- reinforcement learning algorithms
- classifier training
- learning algorithm
- support vector
- partially observable domains
- training process
- knowledge transfer
- data sets
- multi layer perceptron
- markov decision processes
- discriminative classifiers
- artificial neural networks
- trained classifiers
- training set size
- multi agent
- temporal abstractions
- model free
- semi supervised learning
- svm classifier
- optimal control
- classification algorithm