Spatial sequence learning is a complex cognitive process that enables animals and humans to reliably navigate between different locations in a specific order. The goal of this project is to provide a better understanding of how spatial sequence navigation is learned and optimized by integrating information obtained from experimental studies in rats with computational models and autonomous mobile robots.
Spatial sequence learning has been shown to involve brain areas including the hippocampus and the prefrontal cortex (PFC). Recent studies in the rat have shown that neurons in these two areas spontaneously re-activate in short sequences. Much attention has been paid to reactivation during sleep in the context of long-term memory consolidation. The focus of this project is on the role of replay during the awake state, as the animal is learning across multiple trials during the same session. The hypothesis is that the generation of these short sequences of activity in hippocampus allows for global spatial sequence learning in the PFC. The proposed work involves the development of an integrated model of the hippocampus-PFC network that is able to form spatial navigation sequences incorporating: 1) a replay-driven model for memory formation in the hippocampus and 2) a model of spatial sequence learning in the PFC that uses what is known as reservoir computing. The PFC reservoir will consist of large pools of interconnected neural elements that process information dynamically through reverberations. It will consolidate hippocampal replay sequences into larger spatial sequences that may be later recalled by subsets of the original sequences. The proposed work is expected to generate a new mechanistic understanding of the role of replay in memory acquisition in complex tasks such as sequence learning. That understanding will be leveraged and tested on robotic platforms. Original contributions of the proposed work include 1) the use of hippocampal replay to create small chunks of valid trajectories, 2) the use of reservoir computing to learn spatial sequences using the outputs of the hippocampus model, 3) constraining and testing of the model using electrophysiological data in behaving rats and 4) the use of the resulting model in the embodied-cognitive framework of a robot.
A companion project is being funded by the French National Research Agency (ANR).
