Saved in:
Bibliographic Details
Main Author: Perfilev, Vladimir
Format: Recurso digital
Language:English
Published: Zenodo 2026
Subjects:
Online Access:https://doi.org/10.5281/zenodo.19338035
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866902237317431296
author Perfilev, Vladimir
author_facet Perfilev, Vladimir
contents <p>We propose shifting neural network optimization from the parameter space into a dual space of dimensionality. Our core mechanism is the M-matrix—an analytical per-layer Jacobian derived from forward-pass activations, yielding exact gradients without a global computational graph. In this dual formulation, every weight update is the exact optimum of a least-squares problem, solved via a matrix-free Conjugate Gradient (CG) method.<br><br>Code: <a title="Exact Dual-Space Solver" href="https://github.com/v-perfilev/exact-dual-space-solver" rel="noopener">GitHub</a></p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19338035
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Exact Dual-Space Optimization for Neural Networks
Perfilev, Vladimir
Dual-space optimization
M-chain
BCG
FCG
M-grad
Layer parallel architecture
<p>We propose shifting neural network optimization from the parameter space into a dual space of dimensionality. Our core mechanism is the M-matrix—an analytical per-layer Jacobian derived from forward-pass activations, yielding exact gradients without a global computational graph. In this dual formulation, every weight update is the exact optimum of a least-squares problem, solved via a matrix-free Conjugate Gradient (CG) method.<br><br>Code: <a title="Exact Dual-Space Solver" href="https://github.com/v-perfilev/exact-dual-space-solver" rel="noopener">GitHub</a></p>
title Exact Dual-Space Optimization for Neural Networks
topic Dual-space optimization
M-chain
BCG
FCG
M-grad
Layer parallel architecture
url https://doi.org/10.5281/zenodo.19338035