Abstract

What dynamics characterize the transformation of memories over time? Here we introduce a neural network formalization that reveals that memory representations undergo a transition from highly segregated to richly integrated network forms, driven by a combination of neural network reactivations, spreading, and synaptic plasticity rules. Modularity, as a fundamental organizing principle, allows information segregation into cohesive modules, preserving specific sets of information while facilitating efficient spread throughout the network. Through our modeling approach, we reveal an optimal window during this transformation where memories are most susceptible to malleability, suggesting a non-linear or inverted U-shaped function in memory evolution. The results of our model integrate a wide range of experimental phenomena along with accounts of memory consolidation and reconsolidation, offering a unique perspective on memory evolution by leveraging simple architectural neural network property rules.