On the stability of learning
Abstract
Learning algorithms are usually analyzed under the assumption that informative gradients are continuously available. But this is not always true---learning signals in biological neural networks, for example, may be sparse, local, and only intermittently available at any given synapse. In this work, we show that if gradient information is noisy and only intermittently available, learning may be unstable. In particular, even if average learning dynamics converge to the correct optimum, variance in learning may be asymptotically infinite. We explore this point in a number of analytically tractable settings, including single-layer and multi-layer linear networks, and derive explicit conditions under which learning is stable. These conditions complement, but are conceptually distinct from, existing conditions related to the stability of stochastic gradient descent. Our work demonstrates that unbiased credit assignment is not enough for reliable learning; sufficiently frequent and informative feedback is also required, which yields an additional constraint on models of biologically plausible learning.
Related
- Clinical-Anchored Incremental Survival Learning: A Stability-Constrained Framework for Molecularly Enhanced Survival Prediction in Lung Adenocarcinoma
- Analysis of immediate upstream gene regions in orthopoxvirus genomes through the use of DNA Duplex Stability and explainable machine learning
- Temporal stability for individual differences in both habit learning and expression
- Integrating machine learning and statistical techniques to evaluate postural stability during hypoxia
- Mood stability and change map onto distinct insula-centered networks
BibTeX
@misc{vastola2026stability,
title = {{On the stability of learning}},
author = {John J. Vastola},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.35856},
eprint = {2610.35856},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.35856},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - On the stability of learning
AU - John J. Vastola
PY - 2026
DA - 2026/10/26
PB - arXiv
DO - 10.48550/arxiv.2610.35856
UR - https://arxiv.org/abs/2610.35856
AB - Learning algorithms are usually analyzed under the assumption that informative gradients are continuously available. But this is not always true---learning signals in biological neural networks, for example, may be sparse, local, and only intermittently available at any given synapse. In this work, we show that if gradient information is noisy and only intermittently available, learning may be unstable. In particular, even if average learning dynamics converge to the correct optimum, variance in learning may be asymptotically infinite. We explore this point in a number of analytically tractable settings, including single-layer and multi-layer linear networks, and derive explicit conditions under which learning is stable. These conditions complement, but are conceptually distinct from, existing conditions related to the stability of stochastic gradient descent. Our work demonstrates that unbiased credit assignment is not enough for reliable learning; sufficiently frequent and informative feedback is also required, which yields an additional constraint on models of biologically plausible learning.
ER -