Abstraction for Deep Reinforcement Learning

2022/02/10 by Murray Shanahan, Melanie Mitchell, Shanahan, Murray +1
Computer Science · #Neural Networks and Applications #Evolutionary Algorithms and Applications #Computability, Logic, AI Algorithms

paper · pdf · doi:10.48550/arxiv.2202.05839

Abstract

We characterise the problem of abstraction in the context of deep reinforcement learning. Various well established approaches to analogical reasoning and associative memory might be brought to bear on this issue, but they present difficulties because of the need for end-to-end differentiability. We review developments in AI and machine learning that could facilitate their adoption.

Related

• Abstraction and analogy‐making in artificial intelligence
• Abstractions of General Reinforcement Learning
• Deep Reinforcement Learning
• Explainable Reinforcement Learning: A Survey
• Deep Reinforcement Learning