Modular Lifelong Reinforcement Learning via Neural Composition.
Jorge A. MendezHarm van SeijenEric EatonPublished in: ICLR (2022)
Keyphrases
- reinforcement learning
- fitted q iteration
- network architecture
- neural network
- function approximation
- learning algorithm
- state space
- model free
- multi agent
- modular structure
- sensory inputs
- neural model
- st century
- dynamic programming
- learning process
- temporal difference
- data sets
- markov decision processes
- modular neural networks
- web services composition
- biologically plausible
- function approximators
- machine learning
- learning activities
- optimal control
- optimal policy
- action selection
- computational intelligence
- supervised learning
- bio inspired
- neural computation
- associative memory
- learning scenarios