Deep RL agents on evidence-accumulation task mirror rodent behavioral findings

neuroecology · x · 2026-11-21

The author trained deepRL networks on the accumulating tower task and found striking similarities to the rodent literature: history bias, states that switch between sessions, value ramps by evidence and steps. He argues the worm connectome is too underconstrained (given ubiquitous extrasynaptic neuromodulation) for this purpose; more interesting to run networks alone and see where convergences arise.

Original post →

More from Research

Research channel →