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.
@article{arxiv.2202.05839,
title = {Abstraction for Deep Reinforcement Learning},
author = {Murray Shanahan and Melanie Mitchell},
journal= {arXiv preprint arXiv:2202.05839},
year = {2022}
}