Skip to contentSkip to filters

We use cookies for analytics.

Vector production via mental navigation in the entorhinal cortex

Neuroscience ePoster by Sujaya Neupane, Ila Fiete and Mehrdad Jazayeri

Part of COSYNE 2023 posters, Massachusetts Institute of Technology (MIT); Brain and Cognitive Sciences; Massachusetts Institute of Technology (MIT)

Friday

COSYNE 2023
Montreal, Canada
Related content

Abstract

Cognitive maps are hypothesized memory systems that organize sensory experience into knowledge that can be used flexibly in novel scenarios, such as finding new routes during navigation. Evidence for this hypothesis has been largely limited to experiments with fully observable sensory information. However, to serve as a genuine internal source of knowledge, cognitive maps should also be able to support mental computations in the absence of external cues. To test this hypothesis, we developed a mental navigation task in which monkeys must produce vectors that correspond to moving between pairs of randomly selected start and target positions along a fixed and previously experienced sequence of landmarks without visual feedback. Animals succeeded in this task and generalized to unseen landmark pairs, suggesting that they relied on a cognitive map of the landmark sequence. We next recorded from the entorhinal cortex (EC), as a hypothesized substrate for the cognitive map. We found that neurons in a subregion of EC were task modulated and exhibited a periodicity matching the inter-landmark interval. Some EC neurons additionally showed ramping activity with distance-dependent end states, consistent with path integration. We found that neuron pairs with high periodicity scores showed trial- and context-invariant cross-correlation structure, a signature of grid cell continuous attractor network (CAN) states. We thus built a CAN model of entorhinal grid cells augmented with a Hebbian learning mechanism for memorizing the landmark sequence. Attractor states in the model constrained mental navigation such that internally learned landmark inputs transiently slowed the dynamics, acted as local resets, and reduced the overall variability of path integration. Consistent with the predictions from the CAN model, animals’ behavioral variability was captured by a dynamic process punctuated by landmark-induced reset events. Our work thus connects the neural phenomenon of EC periodic neurons to the ability to perform mental navigation.

Topics

Details

Session
Poster Session I