New research reveals a duality between neural network weights and neuron activities that enables a geometric decomposition of the generalization gap. The framework provides a way to interpret the effects of regularization schemes such as stochastic gradient descent and dropout on generalization — and to improve upon these methods.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 digital issues and online access to articles
$119.00 per year
only $9.92 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
References
Neyshabur, B., Tomioka, R. & Srebro, N. Preprint at https://doi.org/10.48550/arXiv.1412.6614 (2014).
Zhang, C., Bengio, S., Hardt, M., Recht, B. & Vinyals, O. In Int. Conf. Learning Representations (ICLR) 2017 https://openreview.net/forum?id=Sy8gdB9xx (2022).
Feng, Y., Zhang, W. & Tu, Y. Nat. Mach. Intell. 5, 908–918 (2023).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Competing interests
The author declares no competing interests.
Rights and permissions
About this article
Cite this article
Gromov, A. Deconstructing the generalization gap. Nat Mach Intell 5, 1340–1341 (2023). https://doi.org/10.1038/s42256-023-00766-7
Published:
Issue Date:
DOI: https://doi.org/10.1038/s42256-023-00766-7