Loss Landscape Characterization of Neural Networks without Over-Parametrization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Islamov, Rustem, Ajroldi, Niccolò, Orvieto, Antonio, Lucchi, Aurelien
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916453522866176
author Islamov, Rustem
Ajroldi, Niccolò
Orvieto, Antonio
Lucchi, Aurelien
author_facet Islamov, Rustem
Ajroldi, Niccolò
Orvieto, Antonio
Lucchi, Aurelien
contents Optimization methods play a crucial role in modern machine learning, powering the remarkable empirical achievements of deep learning models. These successes are even more remarkable given the complex non-convex nature of the loss landscape of these models. Yet, ensuring the convergence of optimization methods requires specific structural conditions on the objective function that are rarely satisfied in practice. One prominent example is the widely recognized Polyak-Lojasiewicz (PL) inequality, which has gained considerable attention in recent years. However, validating such assumptions for deep neural networks entails substantial and often impractical levels of over-parametrization. In order to address this limitation, we propose a novel class of functions that can characterize the loss landscape of modern deep models without requiring extensive over-parametrization and can also include saddle points. Crucially, we prove that gradient-based optimizers possess theoretical guarantees of convergence under this assumption. Finally, we validate the soundness of our new function class through both theoretical analysis and empirical experimentation across a diverse range of deep learning models.
format Preprint
id arxiv_https___arxiv_org_abs_2410_12455
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Loss Landscape Characterization of Neural Networks without Over-Parametrization
Islamov, Rustem
Ajroldi, Niccolò
Orvieto, Antonio
Lucchi, Aurelien
Machine Learning
Optimization and Control
Optimization methods play a crucial role in modern machine learning, powering the remarkable empirical achievements of deep learning models. These successes are even more remarkable given the complex non-convex nature of the loss landscape of these models. Yet, ensuring the convergence of optimization methods requires specific structural conditions on the objective function that are rarely satisfied in practice. One prominent example is the widely recognized Polyak-Lojasiewicz (PL) inequality, which has gained considerable attention in recent years. However, validating such assumptions for deep neural networks entails substantial and often impractical levels of over-parametrization. In order to address this limitation, we propose a novel class of functions that can characterize the loss landscape of modern deep models without requiring extensive over-parametrization and can also include saddle points. Crucially, we prove that gradient-based optimizers possess theoretical guarantees of convergence under this assumption. Finally, we validate the soundness of our new function class through both theoretical analysis and empirical experimentation across a diverse range of deep learning models.
title Loss Landscape Characterization of Neural Networks without Over-Parametrization
topic Machine Learning
Optimization and Control
url https://arxiv.org/abs/2410.12455