Towards Governing Agent's Efficacy: Action-Conditional $β$-VAE for Deep Transparent Reinforcement Learning.
John YangGyuejeong LeeSimyung ChangNojun KwakPublished in: ACML (2019)
Keyphrases
- action selection
- reinforcement learning
- state action
- agent learns
- reward shaping
- action space
- agent receives
- multi agent
- action selection mechanism
- partially observable
- practical reasoning
- state space
- temporal difference
- partial observations
- reward signal
- autonomous learning
- partially observable domains
- joint action
- transition model
- multi agent environments
- learning agent
- markov decision processes
- state abstraction
- intelligent agents
- markov decision process
- function approximation
- multi agent systems
- mobile robot
- reinforcement learning algorithms
- function approximators
- evaluation function
- decision making
- multiagent systems
- sensory inputs
- learning algorithm
- expected reward
- robocup soccer
- agent technology
- agent model
- decision theoretic
- multiple agents
- model free
- plan execution
- internal state
- reward function
- learning capabilities
- autonomous agents
- agent architecture
- external events
- sensing actions
- multiagent reinforcement learning
- stochastic games
- exploration strategy
- learning agents
- action models
- epistemic states