Building cell assembly based associative memory with spiking neurons

Associative memory is a fundamental component of cognitive function. The CA (CA) frame-work, first proposed by Hebb, provides a biologically plausible foundation for understanding the operation of associative memory. The formation and storage of memories depends on structural changes and modifications in synaptic conditions and strength between neurons. This thesis explores networks of spiking neurons to implement CAs and to simulate cognitive functions. The Stroop test, a prominent cognitive interference task, is replicated in a task-completion simulation using binary CAs. Additionally, a question-answering system with CA-based memories models the natural responses predicted by Collin’s hierarchical structure of semantic memory. These experiments underscore the pivotal role of CAs in advancing our understanding of cognitive processes. The thesis also investigates how to improve CA models by proposing standards for evaluating CAs and defining expectations for robust short-term memory simulations. Existing and novel topological structures, incorporating insights and constraints derived from neuroanatomical evidence, are assessed to examine particular CA topologies. A significant contribution is the development of CA models composed solely of excitatory neurons, which exhibit persistent yet non-binary group firing and self-termination behaviours. The small-world rich-get-richer network emerges as a good structure to support CA functions. An associative memory model consisting of a CA group with multiple orthogonal sub-CAs demonstrates that stimulating two of three connected sub-CAs activates the third (2/3 assembly). Moreover, a single sub-CA can participate in multiple 2/3 pairs, enabling the network to achieve a storage capacity of O(N). The phasic interactions among sub-CAs are also explored to support more complex co-activation conditions and demonstrate basic CA behaviours.

Name dc:type.qualificationname

PhD

Level dc:type.qualificationlevel

PhD thesis

Grantor dc:publisher.institution

Middlesex University

Year dc:date.issued

2025

Author dc:creator

• Ji, Y.

Identifier

oai:repository.mdx.ac.uk:27z105

OAI identifier oai:identifier

oai:repository.mdx.ac.uk:27z105