1-dimensional topological invariants to estimate loss surface non-convexity

DOI10.1134/S1064562423701569MaRDI QIDQ6124394zbMATH OpenFDO

Authors D. S. Voronkova, Serguei Barannikov, E. V. Burnaev

Publication date 27 March 2024

Published in Doklady Mathematics (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1134/S1064562423701569

Mathematics Subject Classification ID

General theory of linear operators (47Axx) Differential topology (57Rxx) Computer science support for mathematical research and practice (68Vxx)

Recommendations

• Barcodes as summary of loss function topology
• Visualizing high-dimensional loss landscapes with Hessian directions
• Optimization Landscape of Neural Networks
• Functional loops: monitoring functional organization of deep neural networks using algebraic topology
• Spurious valleys in one-hidden-layer neural network optimization landscapes

Cites work

• scientific article; zbMATH DE number 732198 (Why is no real title available?)
• Precise Arrhenius law for \(p\)-forms: the Witten Laplacian and Morse-Barannikov complex

Cited in

(4)

• Barcodes as summary of loss function topology
• Functional loops: monitoring functional organization of deep neural networks using algebraic topology
• Visualizing high-dimensional loss landscapes with Hessian directions
• Loss barcode: a topological measure of escapability in loss landscapes

This page was built for publication: 1-dimensional topological invariants to estimate loss surface non-convexity

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6124394)