A physical model for pattern completion of highly overlapping patterns for Human Episodic Memory

Patterns of neural activity in the hippocampus change very slowly, perhaps in a scale-invariant manner.
Episodic memory is a consequence of retrieval of a pattern of activity corresponding to a specific event in the past.
The physics of attractor networks have been extensively studied. However, it is well known that attractor networks have a small capacity when storing the correlated patterns. This coupled with the strong overlap in patterns of neural activity in the hippocampus makes attractor networks ill-suited to describe human episodic memory.
Using a concise description of hippocampal time cells, we develop a formal model for pattern completion of hippocampal representations. The model overcomes pattern similarity by relying on the coupling between different temporal scales. Because the model uses the recency of the desired memory as part of the retrieval process, it can be understood as an ``address-addressable’’ memory. In this sense, time serves as a scaffolding to organize different experiences. We study the capacity and dynamics of this model for memory retrieval and compare to properties of human episodic memory.