A Neuro-Symbolic Classification Algorithm Using Neural Cell Assemblies

Florin Leon

Abstract: This paper introduces a neuro-symbolic classification algorithm inspired by the SUSTAIN model in cognitive psychology. The proposed algorithm is implemented using neural cell assemblies, which can be seen as an intermediate level between individual neurons and higher-level cognitive functions. They offer the advantages of neural representations but can also be interpreted symbolically. Two methods are introduced to transform inputs into cell assembly representations. The algorithm creates prototypes from training instances and automatically estimates the weights of the problem dimensions. Although prototypes are typically used to classify new instances based on similarity, the paper also suggests a method to extract explicit, simple rules from the prototypes, encoded using the same cell assembly representation.