Reward-agnostic Fine-tuning: Provable Statistical Benefits of Hybrid Reinforcement Learning.
Gen LiWenhao ZhanJason D. LeeYuejie ChiYuxin ChenPublished in: CoRR (2023)
Keyphrases
- fine tuning
- reinforcement learning
- viable alternative
- function approximation
- state space
- fine tune
- model free
- eligibility traces
- learning algorithm
- statistical models
- statistical analysis
- optimal policy
- transfer learning
- information theoretic
- markov decision processes
- statistical information
- hybrid learning
- average reward
- learning agent
- fine tuned
- policy search
- partially observable
- reinforcement learning algorithms
- learning classifier systems
- action selection
- machine learning
- confidence intervals
- data driven
- supervised learning
- multi agent
- temporal difference
- learning problems
- markov decision process
- statistical methods
- action space
- domain specific
- general purpose
- policy gradient
- lower bound
- partially observable environments