[PS-2.4] The schema theory of memory consolidation: A stochastic neural network model

Shippi, M. 1, 2, 3 , van Rossum, M. . 1 , G.M. Morris, R. 2 & Lansner, A. 3

1 Institute for Adaptive and Neural Computation, School of Informatics, University of Edinburgh, Edinburgh, UK
2 Centre for Cognitive and Neural Systems, University of Edinburgh, Edinburgh, UK
3 Department of Computational Biology, KTH Royal Institute of Technology, Stockholm, Sweden

Systems memory consolidation has been thought as a long reorganization process, taking weeks or longer, whereby hippocampus dependent memory traces of information become stabilized in the neocortex, and some of them become hippocampus independent. Theoretical studies suggest that a long gradual consolidation process is important in order to avoid catastrophic interference (McLelland et al, Psych. Rev., 1995). However, based on experimental studies, we have shown that memory consolidation can occur rapidly without interference, if an associative framework of knowledge has been previously created, called an associative ?schema?, into which new memory traces can be incorporated (Tse, Langston et al, Science, 2007). This schema-specific rapid consolidation is accompanied, at the time of memory encoding, by an upregulation of plasticity-related immediate early genes in the prelimbic region of medial prefrontal cortex (Tse, Takeuchi et al, Science, 2011). Pharmacological interventions targeted at the prelimbic region prevented both, new learning and the recall of remotely, and recently consolidated information, complementing other studies (Lesburgueres et al, Science, 2011). These data shed light on the neural mechanisms of rapid assimilation into memory schemas which challenge current theoretical frameworks of memory consolidation, such as the complementary learning theory of distinct fast hippocampal and slow cortical learning systems (McLelland et al, Psych. Rev., 1995).

Using Restricted Boltzman Machines, we present a theoretical model of systems memory consolidation including the concept of schema which explicitly considers the role of prior knowledge in guiding the consolidation process. We investigate the role of the hippocampus, the prelimbic cortex and their dynamic interaction during schema learning, and memory encoding and assimilation of additional related memory traces. Our results are in agreement with both human (van Kesteren et al, PNAS, 2010) and animal (Tse, Takeuchi et al, Science, 2011, Tse, Langston et al, Science, 2007) studies of schemas and memory consolidation.