Reset & Distill: A Recipe for Overcoming Negative Transfer in Continual Reinforcement Learning.
Hongjoon AhnJinu HyeonYoungmin OhBosun HwangTaesup MoonPublished in: CoRR (2024)
Keyphrases
- multi agent
- reinforcement learning
- transfer learning
- positive and negative
- knowledge transfer
- function approximation
- state space
- control problems
- machine learning
- reinforcement learning algorithms
- learning tasks
- markov decision processes
- dynamic programming
- temporal difference
- optimal policy
- labeled data
- collaborative filtering
- supervised learning
- learning process
- cross domain
- search space
- data sets