Neural Networks as Local-to-Global Computations

Vicente Bosca, Robert Ghrist · 2026

Read Neural Networks as Local-to-Global Computations by Vicente Bosca, Robert Ghrist (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-2603-14831.txt.

Description

Bosca and Ghrist rebuild feedforward ReLU networks as cellular sheaves with local-to-global flow. Each forward-pass quantity becomes a vertex in the sheaf. Affine steps, activations and outputs become restriction maps on edges. The coboundary on free coordinates is unitriangular, with determinant one. Relative cohomology vanishes under that shape. The forward output is thus the unique harmonic extension of input data. A sheaf heat equation settles to that output exponentially. Pinned neurons and local training without backpropagation follow from the frame.

Catalog record

CREATORVicente Bosca, Robert Ghrist
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE148767
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.14831

Browse the full library · About the Tower of Babel