A neural model of short-term and intermediate-term memory using neuro-symbolic representations for semantic and spatial memory tasks

Proceedings of the Annual Meeting of the Cognitive Science Society, 2026

Jakeb Chouinard, Chris Eliasmith

Abstract

Memory is a vital part of cognition. Its many parts determine how one learns, plans, and ultimately experiences the world around them. As a bridge between short-term memory and long-term memory, intermediate-term memory allows for semi-stable memories to be rapidly encoded through short-term synaptic plasticity. We present a model of short-term and intermediate-term memory based around Semantic Pointers and realize the model in a spiking neural network. Unlike past work, we characterize short-term memory as a dynamical system and realize intermediate-term memory as implicit association learning. Further, we demonstrate our model’s ability to generalize across domains by achieving human-like results on semantic and spatial memory tasks with minimal parameter changes. By examining Cohen’s ℎ values first within and then between tasks, we find an overall small difference (Mean:|\bar (ℎ)| = 0.12 ± 0.06, Median: |\hat (ℎ)| = 0.16), suggesting strong similarity between model and human data across all three tasks.

Full text links

 PDF

Conference Proceedings

Booktitle
Proceedings of the Annual Meeting of the Cognitive Science Society
Volume
48
Month
07
Note
In press

Cite

Plain text

BibTeX