Cortical networks can maintain memories for decades, despite short lifetime of synaptic strength. Can a neural network store long-lasting memories in unreliable synapses? Here we study the effects of random noise on the stability of memory stored in synapses of an attractor neural network. The model includes ongoing spike timing dependent plasticity (STDP). We show that certain class of STDP rules can lead to stabilization of memory patterns stored in the network. The stabilization results from rehearsals induced by noise. We show that unstructured neural noise, after passing through the recurrent network weights, carries the imprint of all of the memory patterns in temporal correlations. Under certain strict conditions, STDP combined with these correlations can lead to reinforcement of all of the existing patterns, even those that are never explicitly visited, i.e. unused. We show that stabilization of unused memories occurs for asymmetric STDP learning rules (Figure 1), while symmetric non-negative rules do not have this property. Thus, we propose that, unstructured neural noise can stabilize the existing structure of synaptic connectivity. Our findings may provide the functional reason for highly irregular spiking displayed by cortical neurons and provide justification for models of system memory consolidation. Out theory makes experimentally testable predictions, such as that synaptic strengths in the cortex should be correlated with the correlations in the pre- and postsynaptic neural activity on the synapse-by-synapse basis. We thus propose that unreliable neural activity is the feature that helps cortical networks maintain stable connections.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.