Off-line memory reprocessing following on-line unsupervised learning strongly improves recognition performance in a hierarchical visual memory

Recently, experience-driven unsupervised learning was shown to create combinatorial parts-based representations in a model of hierarchical visual memory. Examining the memory´s ability to recognize persons from a database of natural face images, we show that an off-line, sleep-like operating regime of the memory domain results in a significant improvement of the system´s ability to generalize over novel face views. Surprisingly, the positive effect turns out to be independent of synapse-specific plasticity, relying entirely on a homeostatic mechanism equalizing the intrinsic excitability levels of the units within the memory network. We show that this excitability equalization is the main cause for the improvement of memory function. A possible relation to cortical off-line memory reprocessing during certain sleep stages is discussed.