Modeling memory and learning with molecular dynamics

Abstract

The present study transfers molecular dynamics (MD) in condensed matter physics to simulate human memory. In our MD model, a spatial representation of memory of an object is identified where constituent elements of the object are depicted in terms of physical particles. Learning is implemented as attractive forces between particles, which drive particles to gather gradually together and form closely packed clusters eventually. With this, the robustness of the memory is elucidated by the total distance between particles within the cluster. Commonly, a well-established memory of an object corresponds to a cluster with small inter-particle distances. As a demonstration, the present MD model captures a set of interference effects, including retroactive interference, proactive interference and interference between concurrently learned materials that are similar to each other. Different from the existing computational models, our MD model, relying essentially on only one parameter of the force between particles, may be useful to simulate and understand diverse phenomena associated with memory and learning.