Evaluating Loss Landscapes from a Topology Perspective

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Xie, Tiankai, Geniesse, Caleb, Chen, Jiaqing, Yang, Yaoqing, Morozov, Dmitriy, Mahoney, Michael W., Maciejewski, Ross, Weber, Gunther H.
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866917838638284800
author Xie, Tiankai
Geniesse, Caleb
Chen, Jiaqing
Yang, Yaoqing
Morozov, Dmitriy
Mahoney, Michael W.
Maciejewski, Ross
Weber, Gunther H.
author_facet Xie, Tiankai
Geniesse, Caleb
Chen, Jiaqing
Yang, Yaoqing
Morozov, Dmitriy
Mahoney, Michael W.
Maciejewski, Ross
Weber, Gunther H.
contents Characterizing the loss of a neural network with respect to model parameters, i.e., the loss landscape, can provide valuable insights into properties of that model. Various methods for visualizing loss landscapes have been proposed, but less emphasis has been placed on quantifying and extracting actionable and reproducible insights from these complex representations. Inspired by powerful tools from topological data analysis (TDA) for summarizing the structure of high-dimensional data, here we characterize the underlying shape (or topology) of loss landscapes, quantifying the topology to reveal new insights about neural networks. To relate our findings to the machine learning (ML) literature, we compute simple performance metrics (e.g., accuracy, error), and we characterize the local structure of loss landscapes using Hessian-based metrics (e.g., largest eigenvalue, trace, eigenvalue spectral density). Following this approach, we study established models from image pattern recognition (e.g., ResNets) and scientific ML (e.g., physics-informed neural networks), and we show how quantifying the shape of loss landscapes can provide new insights into model performance and learning dynamics.
format Preprint
id arxiv_https___arxiv_org_abs_2411_09807
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Evaluating Loss Landscapes from a Topology Perspective
Xie, Tiankai
Geniesse, Caleb
Chen, Jiaqing
Yang, Yaoqing
Morozov, Dmitriy
Mahoney, Michael W.
Maciejewski, Ross
Weber, Gunther H.
Machine Learning
Artificial Intelligence
Characterizing the loss of a neural network with respect to model parameters, i.e., the loss landscape, can provide valuable insights into properties of that model. Various methods for visualizing loss landscapes have been proposed, but less emphasis has been placed on quantifying and extracting actionable and reproducible insights from these complex representations. Inspired by powerful tools from topological data analysis (TDA) for summarizing the structure of high-dimensional data, here we characterize the underlying shape (or topology) of loss landscapes, quantifying the topology to reveal new insights about neural networks. To relate our findings to the machine learning (ML) literature, we compute simple performance metrics (e.g., accuracy, error), and we characterize the local structure of loss landscapes using Hessian-based metrics (e.g., largest eigenvalue, trace, eigenvalue spectral density). Following this approach, we study established models from image pattern recognition (e.g., ResNets) and scientific ML (e.g., physics-informed neural networks), and we show how quantifying the shape of loss landscapes can provide new insights into model performance and learning dynamics.
title Evaluating Loss Landscapes from a Topology Perspective
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2411.09807