Self-generation of reward in reinforcement learning by universal rules of interaction with the external environment.
Kentarou KurashigeKaoru NikaidoPublished in: RiiSS (2014)
Keyphrases
- reinforcement learning
- agent learns
- learning agent
- association rules
- state space
- learning classifier systems
- complex environments
- initially unknown
- function approximation
- agent environment
- real time
- reinforcement learning algorithms
- exploration strategy
- markov decision processes
- association rule mining
- human computer interaction
- user interaction
- dynamic programming
- model free
- learning algorithm
- sensory inputs
- multi agent environments
- eligibility traces
- internal state
- partially observable environments
- interaction model
- temporal difference
- human interaction
- production rules
- rule sets
- decision rules
- virtual world
- mobile robot
- machine learning