Assessing the role of hippocampal replay in retrospective revaluation

Summary

Date: 
December 12, 2018 - 12:00pm - 1:00pm
Location: 
Northwest Building, Room 243
About the Speaker
Name: 
Hector Penagos-Vargas
Speaker Title: 
Postdoctoral Fellow
Speaker Affiliation: 
Gershman Lab & Wilson Lab (MIT)

Computational accounts of decision-making portray action selection as the result of a competition between model-free and model-based systems. The former relies on repeatedly selecting previously rewarded actions while the latter uses an internal model of the environment to plan goal-directed behavior. The model-based system thus reflects the cognitive map hypothesis. Recent human data suggest that in the case of retrospective revaluation tasks, where subjects are allowed to change their preference between two choices after new independent evidence has been presented, a cooperative interaction between the two systems better accounts for the observed decision-making behavior. A key prediction of this cooperative architecture is that the model-based system replays and simulates experiences offline, linking all available evidence, to train the model-free system for future online action selection.

Given the strong resemblance between hippocampal replay and the predicted role of the model-based system in retrospective revaluation, we designed a navigational version of the human task and recorded spiking activity in the hippocampus of four rats. We show that, like humans, rats are able to change prior preferences in light of new evidence with a stronger effect at the beginning of the testing sessions. Inspection of hippocampal activity during periods of quiet wakefulness reveals that the hippocampus progressively incorporates new independent information, and represents it along previously acquired experiences. The joint representation of accumulated evidence is consistent with an offline mechanism that evaluates all potential options subjects could consider for action selection. Together with previous findings showing that pharmacological inactivation of the hippocampus prevents behavior consistent with model-based planning in rats, these results suggest that hippocampal replay is a suitable mechanism for the extraction and evaluation of the reward structure of the environment that allows model-based planning in line with the cognitive map hypothesis.