Back to search

University of Southampton

Biologically inspired neuromorphic algorithms for low power learning

Abstract

dc:description.abstract

While neuromorphic computing offers a potential low power alternative to current resource-hungry AI, it requires fresh algorithmic approaches. The field needs a new generation of Spiking Neural Networks (SNNs) that have not evolved with conventional hardware but instead align with characteristics of nascent neuromorphic platforms. With this goal in mind, we research biologically inspired networks that expand by incorporating fresh neurons to facilitate continual learning. This architectural plasticity is largely overlooked as a technique in conventional machine learning and is inspired by biological adult neurogenesis.<br/><br/>To underpin our research, we provide a unifying definition and mathematical formulation for a group of associative memory networks that expand during learning. Our formulation enables theoretical network optimisation which demonstrates greater than 10,000 times improvement in memory capacity compared to previous work. We then direct our focus on the learning step itself and propose several new neurogenesis learning strategies that increase in memory recall accuracy from 69% to 96%, while retaining a fixed resource budget. Furthermore, through application of our findings to the more complex task of Class Incremental Learning (CIL), we are able demonstrate that dedicating separate resources to the independent learning of each class can help to regulate learning.<br/><br/>Our consolidated findings include: methods to increase accuracy by reducing bias towards specific memories during learning; methods to reduce the computational cost of architectural plasticity during learning; and empirical demonstration that a modular network topology can be exploited to optimise neurogenesis. In summary, we extend existing neuromorphic research through the formal articulation, assessment, and advancement of learning strategies that incorporate architectural plasticity. The research responds directly to calls from the neuromorphic research community for original approaches to progress neuromorphic algorithm development.<br/>

Degree

thesis:*
Name dc:type.qualificationname
Ph.D.
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Southampton
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Warr, Katy
Advisors dc:contributor.advisor
  • Hare, Jonathon
  • Thomas, David B.

Rights

Language dc:language
English, eng

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
University of Southampton
Base URL
eprints.soton.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Warr, Katy. Biologically inspired neuromorphic algorithms for low power learning. doctoral thesis, University of Southampton, 2026. https://doi.org/10.5258/SOTON/01/T037