Search Papers | Poster Sessions | All Posters
Poster B87 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Spatially- and non-spatially tuned hippocampal neurons exhibit distinct linear and non-linear representations
Maxime Daigle1 (), Benjamin Corrigan2, Julio Martinez-Trujillo2, Pouya Bashivan1; 1McGill University, Canada, 2Western University, Canada
The hippocampus is known to flexibly represents spatial and non-spatial features of the environment in a task-dependent manner. However, the underlying neural mechanisms governing this contextual adaptability remain elusive. To investigate this, we trained artificial neural networks (ANNs) to perform a navigation-dependent associative memory task mirroring the one performed by macaque monkeys. Using the unit activities of these models, we constructed predictive models of macaque monkey CA3 neurons and measured which types of model better capture the neural computation within the hippocampus. Our results reveal that spatially-tuned neurons predominantly code linear feature combinations, while non-spatially tuned neurons are better explained by non-linear spatiotemporal feature combinations. Moreover, we show that an ANN trained for a navigation-dependent associative memory task learns non-linear spatiotemporal representations that are substantially more aligned with those in the hippocampus compared to alternative models. Altogether, our results shed light on the nature of selectivity across multiple feature dimensions by revealing that the linear and non-linear mixing of features by distinct hippocampal neurons matches surprisingly well with their tendency to be spatially-tuned or not.
Keywords: Hippocampus Associative memory Navigation