Learning with Shallow Neural Networks on Cluster-Structured Features

Elisabetta Cornacchia, Laurent Massoulié · 2026

Read Learning with Shallow Neural Networks on Cluster-Structured Features by Elisabetta Cornacchia, Laurent Massoulié (2026) 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-2605-14927.txt.

Description

Elisabetta Cornacchia and Laurent Massoulié ask how clustered input features change the difficulty of learning with shallow neural networks. Real inputs like images and genomes carry correlations within the input space, not just in the target function. In their model, targets depend on few latent Boolean variables, and features cluster around those variables. Under an identifiability condition, layerwise gradient descent needs samples scaling with the hidden variables alone, independent of input size at high signal-to-noise.

Catalog record

CREATORElisabetta Cornacchia, Laurent Massoulié
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE99428
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.14927

Browse the full library · About the Tower of Babel