Loss is its own Reward: Self-Supervision for Reinforcement Learning.
Evan ShelhamerParsa MahmoudiehMax ArgusTrevor DarrellPublished in: CoRR (2016)
Keyphrases
- reinforcement learning
- function approximation
- temporal difference
- eligibility traces
- reward function
- markov decision processes
- state space
- multi agent
- reinforcement learning algorithms
- partially observable environments
- learning agent
- reinforcement learning methods
- supervised learning
- dynamic programming
- policy search
- reward shaping
- machine learning
- model free
- optimal policy
- learning process
- active learning
- learning algorithm
- information loss
- action selection
- optimal control
- learning problems
- transfer learning
- markov decision process
- function approximators
- average reward
- markov chain
- total reward
- robotic control
- data sets