Neural stochasticity and criticality in memory replay

Description du projet

Les avalanches pourraient constituer une force positive de la nature pour consolider la mémoire

La formation, le stockage et la récupération des souvenirs sont essentiels à notre vie quotidienne, à notre développement personnel et professionnel, à nos relations interpersonnelles et à notre bien-être général. Si ces fonctions de la mémoire font défaut, les conséquences peuvent être dévastatrices, comme en témoignent la démence ou d’autres types de troubles ou de pertes de la mémoire. L’ensemble du processus de formation, de stockage et de récupération présente des dimensions spatiales et temporelles complexes, car les souvenirs se déplacent d’une partie du cerveau à l’autre durant le processus de consolidation. Le projet BrownianReactivation, financé par l’UE, associe des expériences à la modélisation mathématique pour étudier les «avalanches d’activité» sur la piste des souvenirs, des substrats à court terme (hippocampe) aux substrats à long terme (cortex).

Objectif

Memories are not stored in a stable manner in the brain. Instead they undergo an almost constant process of transformation and reorganization from the very moment of their formation. This phenomenon, going under the name of memory consolidation is also at the core of fundamental changes in the nature of the information carried by the neural representations. In fact, the switch from hippocampal-based short-term memories to cortex-based long-term memories also coincides with the transitions from episodic to semantic memories. In this proposal, we aim at investigating the role played in memory consolidation by stochasticity and criticality of neural dynamics. We will develop a quantitative approach to this question by combining statistical analysis of newly recorded data, with mathematical modelling of the cortical-hippocampal interaction. Our hypothesis takes the emergence of long-range correlation as the central mechanism characterizing semantic memory formation. We propose that cortical neural circuits poised at criticality can exploit activity avalanches associated with this state to induce the collective activation of large portions of the brain necessary to reconstruct the complete input statistics. Using a hierarchical neural network trained with an input of random samplings of the episodic representations stored in the hippocampus, we will explore the interaction between the performance of the network and its closeness to a critical state. We will complement this work with an analysis of the presence of criticality in the cortical cell assemblies formed after learning a complex behavioural task. We expect these cell assemblies to show maximal contribution to the propagation of neural avalanches and to present consistent coordination with the hippocampal activity. We intend to integrate these results in a coherent theory of long-term memory formation, explaining the ability to generalize starting from event-based experiences.