BriefGPT.xyz
May, 2023
Theta序列作为资格跟踪: 信用分配的生物学解决方案
Theta sequences as eligibility traces: a biological solution to credit assignment
HTML
PDF
Tom M George
TL;DR
使用海马体中的theta序列作为解决策略评估的方丈,可以在没有长期记忆的情况下进行授分,从而压缩行为并扩展短暂的神经记忆痕迹。
Abstract
credit assignment problems
, for example
policy evaluation
in
rl
, often require bootstrapping prediction errors through preceding states \t
→