Modeling dynamics of memory integration and retrieval in the hippocampus

Inspired by the hippocampal CA3–CA1 synaptic network model for memory storage, stored memory engrams are encoded as trajectories in a stable heteroclinic network (SHN), where saddle points represent their information blocks. In this study, the SHN is refined as a hierarchical network by descending the non-zero coordinate values of all saddle points. The nodes of the network are saddle points, and their weights are non-zero coordinate values, indicating the memory strength of blocks. The directed edges are the heteroclinic orbits between saddle points, their weights are the positive eigenvalues, implying the intrinsic excitability of postsynaptic neurons. The integration process is described as the process of inserting new memories to existing hierarchical networks through a repeated information block, while the retrieval process of the memory is described as the orderly access to the nodes corresponding to all the needed information blocks. There is an interference between memories containing identical information blocks, resulting in memories with higher edge weights being more readily retrievable, while unretrieved memories may be obscured, leading to forgetting. Retrieval efficiency and capability are defined to evaluate the integration quality of the hierarchical network, optimizing the process of learning and memory.