Efficient Event-based Delay Learning in Spiking Neural Networks

Balázs Mészáros, James C. Knight, Thomas Nowotny · 2025

Read Efficient Event-based Delay Learning in Spiking Neural Networks by Balázs Mészáros, James C. Knight, Thomas Nowotny (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-07331.txt.

Description

Balazs Meszaros, James Knight, and Thomas Nowotny present a method for teaching delays in spiking neural networks, in a 2025 paper. These networks speak in brief spikes and promise big energy savings over standard networks. A neuron's own memory is short, set by its decay rates. Delays on the links between neurons add a second, longer memory. Tuning those delays by gradient methods is the hard part this paper takes on.

Their method grows from a formalism that gives exact gradients for spike times. It handles neurons that fire more than once and, the authors say, is the first delay method to run on recurrent spiking networks. Tests ranged from a toy sequence task to spoken digit and speech sets, plus Braille letter reading. Learned delays lifted accuracy over no-delay networks, above all in small ones. The method also used under half the memory of the prior best.

Catalog record

CREATORBalázs Mészáros, James C. Knight, Thomas Nowotny
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-07331

Browse the full library · About the Tower of Babel