Geometry of Linear Convolutional Networks

DOI10.1137/21M1441183OpenAlexW3191453487WikidataQ114073981 ScholiaQ114073981MaRDI QIDQ5097687

Matthew Trager, Guido Montúfar, Kathlén Kohn, Thomas Merkh

Publication date: 31 August 2022

Published in: SIAM Journal on Applied Algebra and Geometry (Search for Journal in Brave)

Full work available at URL: https://arxiv.org/abs/2108.01538

zbMATH Keywords

tensorsemi-algebraic setcritical pointgradient flowToeplitz matrixdiscriminantcirculant matrixalgebraic statisticsEuclidean distance degreelinear networkfunction space description of neural networks

Mathematics Subject Classification ID

Artificial neural networks and deep learning (68T07) Semialgebraic sets and related spaces (14P10) Hypersurfaces and algebraic geometry (14J70) Algebraic statistics (62R01) Polynomial optimization (90C23)

Related Items

Side effects of learning from low-dimensional data embedded in a Euclidean space

Uses Software

• Macaulay2

Cites Work

• Unnamed Item
• Some remarks on equations defining coincident root loci
• Duality of multiple root loci
• Geometric reasoning with logic and algebra
• Wasserstein distance to independence models
• Algebraic Statistics
• Optimal Transport to a Variety
• Learning deep linear neural networks: Riemannian gradient flows and convergence to global minimizers
• Mixtures and products in two graphical models
• Maximum likelihood estimation of the Latent Class Model through model boundary decomposition
• Nonconvex Optimization Meets Low-Rank Matrix Factorization: An Overview
• Algorithms in real algebraic geometry
• The Euclidean distance degree of an algebraic variety