dorsal/arxiv
View SchemaWhen Response Variability Increases Neural Network Robustness to Synaptic Noise
| Authors | Gleb Basalyga, Emilio Salinas |
|---|---|
| Categories | |
| ArXiv ID | q-bio/0510036 |
| URL | https://arxiv.org/abs/q-bio/0510036 |
Abstract
Cortical sensory neurons are known to be highly variable, in the sense that responses evoked by identical stimuli often change dramatically from trial to trial. The origin of this variability is uncertain, but it is usually interpreted as detrimental noise that reduces the computational accuracy of neural circuits. Here we investigate the possibility that such response variability might, in fact, be beneficial, because it may partially compensate for a decrease in accuracy due to stochastic changes in the synaptic strengths of a network. We study the interplay between two kinds of noise, response (or neuronal) noise and synaptic noise, by analyzing their joint influence on the accuracy of neural networks trained to perform various tasks. We find an interesting, generic interaction: when fluctuations in the synaptic connections are proportional to their strengths (multiplicative noise), a certain amount of response noise in the input neurons can significantly improve network performance, compared to the same network without response noise. Performance is enhanced because response noise and multiplicative synaptic noise are in some ways equivalent. These results are demonstrated analytically for the most basic network consisting of two input neurons and one output neuron performing a simple classification task, but computer simulations show that the phenomenon persists in a wide range of architectures, including recurrent (attractor) networks and sensory-motor networks that perform coordinate transformations. The results suggest that response variability could play an important dynamic role in networks that continuously learn.
{
"annotation_id": "9274666a-7b70-41f3-972e-ed931086532d",
"date_created": "2026-03-02T18:01:32.284000Z",
"date_modified": "2026-03-02T18:01:32.284000Z",
"file_hash": "99e66db5cbed0b41943460a15e4d5d1d208ae3fca08c9c6df0d4bd70ee8d5ec1",
"private": false,
"record": {
"abstract": "Cortical sensory neurons are known to be highly variable, in the sense that\nresponses evoked by identical stimuli often change dramatically from trial to\ntrial. The origin of this variability is uncertain, but it is usually\ninterpreted as detrimental noise that reduces the computational accuracy of\nneural circuits. Here we investigate the possibility that such response\nvariability might, in fact, be beneficial, because it may partially compensate\nfor a decrease in accuracy due to stochastic changes in the synaptic strengths\nof a network. We study the interplay between two kinds of noise, response (or\nneuronal) noise and synaptic noise, by analyzing their joint influence on the\naccuracy of neural networks trained to perform various tasks. We find an\ninteresting, generic interaction: when fluctuations in the synaptic connections\nare proportional to their strengths (multiplicative noise), a certain amount of\nresponse noise in the input neurons can significantly improve network\nperformance, compared to the same network without response noise. Performance\nis enhanced because response noise and multiplicative synaptic noise are in\nsome ways equivalent. These results are demonstrated analytically for the most\nbasic network consisting of two input neurons and one output neuron performing\na simple classification task, but computer simulations show that the phenomenon\npersists in a wide range of architectures, including recurrent (attractor)\nnetworks and sensory-motor networks that perform coordinate transformations.\nThe results suggest that response variability could play an important dynamic\nrole in networks that continuously learn.",
"arxiv_id": "q-bio/0510036",
"authors": [
"Gleb Basalyga",
"Emilio Salinas"
],
"categories": [
"q-bio.NC",
"cond-mat.dis-nn"
],
"title": "When Response Variability Increases Neural Network Robustness to Synaptic Noise",
"url": "https://arxiv.org/abs/q-bio/0510036"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "98e2ea48-692b-4480-a1c2-5bb7a80e5c75",
"id": "arXiv Dataset IDs",
"type": "Model",
"variant": "snapshot-2026-03-01",
"version": "0.1.0"
},
"user_id": 1000002
}