A neuronal model for the development of schemas and their role in systems memory consolidation

New learning of declarative memories and the consolidation of memories can be facilitated if the newly learned information is consistent with preexisting knowledge. The existence of a knowledge base or “schema” accelerates new learning and enables plasticity during acquisition to occur also outside the hippocampus. However, the neural representation underlying a schema, how it can be used to promote new learning, and the cellular/network mechanisms that allow for the development of a schema are poorly understood. Here we hypothesize that schemas are related to extra-hippocampal neural representations of memory items with particularly high population sparseness, which is theoretically highly efficient for associative learning. However, there is a lack of understanding if and to what degree high population sparseness can be achieved in neural networks. Thus, the aim of the project is to develop and test a mechanistic theory for the development of population-sparse representations of particularly significant memory items.

Archive – B01 (SFB1315/1)

Graphical abstract: Schemas and neural representations. Top: In hippocampus-like networks, representations are sparse. The small size of bright-colored boxes indicates the small fraction f of highly active neurons representing items. Sparse representations allow fast (“one-shot”) learning of an association between two items because the fraction of synapses involved is proportional to f2 (black double-headed arrow). Bottom: Neocortex-like networks with distributed representations (pale colored boxes indicate active neurons) require slow or “interleaved” training to avoid catastrophic interference with other memories. We propose that some representations become sparser through processes related to systems memory consolidation (magenta solid arrows), which allows faster learning and faster consolidation (magenta dashed arrow).