Saved in:
Bibliographic Details
Main Authors: Chou, Hung-Hsu, Maly, Johannes, Verdun, Claudio Mayrink, da Costa, Bernardo Freitas Paulo, Mirandola, Heudson
Format: Preprint
Published: 2022
Subjects:
Online Access:https://arxiv.org/abs/2207.08437
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917948731424768
author Chou, Hung-Hsu
Maly, Johannes
Verdun, Claudio Mayrink
da Costa, Bernardo Freitas Paulo
Mirandola, Heudson
author_facet Chou, Hung-Hsu
Maly, Johannes
Verdun, Claudio Mayrink
da Costa, Bernardo Freitas Paulo
Mirandola, Heudson
contents Over the past years, there has been significant interest in understanding the implicit bias of gradient descent optimization and its connection to the generalization properties of overparametrized neural networks. Several works observed that when training linear diagonal networks on the square loss for regression tasks (which corresponds to overparametrized linear regression) gradient descent converges to special solutions, e.g., non-negative ones. We connect this observation to Riemannian optimization and view overparametrized GD with identical initialization as a Riemannian GD. We use this fact for solving non-negative least squares (NNLS), an important problem behind many techniques, e.g., non-negative matrix factorization. We show that gradient flow on the reparametrized objective converges globally to NNLS solutions, providing convergence rates also for its discretized counterpart. Unlike previous methods, we do not rely on the calculation of exponential maps or geodesics. We further show accelerated convergence using a second-order ODE, lending itself to accelerated descent methods. Finally, we establish the stability against negative perturbations and discuss generalization to other constrained optimization problems.
format Preprint
id arxiv_https___arxiv_org_abs_2207_08437
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Get rid of your constraints and reparametrize: A study in NNLS and implicit bias
Chou, Hung-Hsu
Maly, Johannes
Verdun, Claudio Mayrink
da Costa, Bernardo Freitas Paulo
Mirandola, Heudson
Optimization and Control
Numerical Analysis
Over the past years, there has been significant interest in understanding the implicit bias of gradient descent optimization and its connection to the generalization properties of overparametrized neural networks. Several works observed that when training linear diagonal networks on the square loss for regression tasks (which corresponds to overparametrized linear regression) gradient descent converges to special solutions, e.g., non-negative ones. We connect this observation to Riemannian optimization and view overparametrized GD with identical initialization as a Riemannian GD. We use this fact for solving non-negative least squares (NNLS), an important problem behind many techniques, e.g., non-negative matrix factorization. We show that gradient flow on the reparametrized objective converges globally to NNLS solutions, providing convergence rates also for its discretized counterpart. Unlike previous methods, we do not rely on the calculation of exponential maps or geodesics. We further show accelerated convergence using a second-order ODE, lending itself to accelerated descent methods. Finally, we establish the stability against negative perturbations and discuss generalization to other constrained optimization problems.
title Get rid of your constraints and reparametrize: A study in NNLS and implicit bias
topic Optimization and Control
Numerical Analysis
url https://arxiv.org/abs/2207.08437