Loss Barcode: A Topological Measure of Escapability in Loss Landscapes

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Barannikov, Serguei, Voronkova, Daria, Mironenko, Alexander, Trofimov, Ilya, Korotin, Alexander, Sotnikov, Grigorii, Burnaev, Evgeny
Natura: Preprint
Pubblicazione: 2020
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910039361454080
author Barannikov, Serguei
Voronkova, Daria
Mironenko, Alexander
Trofimov, Ilya
Korotin, Alexander
Sotnikov, Grigorii
Burnaev, Evgeny
author_facet Barannikov, Serguei
Voronkova, Daria
Mironenko, Alexander
Trofimov, Ilya
Korotin, Alexander
Sotnikov, Grigorii
Burnaev, Evgeny
contents Neural network training is commonly based on SGD. However, the understanding of SGD's ability to converge to good local minima, given the non-convex nature of loss functions and the intricate geometric characteristics of loss landscapes, remains limited. In this paper, we apply topological data analysis methods to loss landscapes to gain insights into the learning process and generalization properties of deep neural networks. We use the loss function topology to relate the local behavior of gradient descent trajectories with the global properties of the loss surface. For this purpose, we define the neural network's Topological Obstructions score ("TO-score") with the help of robust topological invariants, barcodes of the loss function, which quantify the escapability of local minima for gradient-based optimization. Our two principal observations are: 1) the loss barcode of the neural network decreases with increasing depth and width, therefore the topological obstructions to learning diminish; 2) in certain situations there is a connection between the length of minima segments in the loss barcode and the minima's generalization errors. Our statements are based on extensive experiments with fully connected, convolutional, and transformer architectures and several datasets including MNIST, FMNIST, CIFAR10, CIFAR100, SVHN, and multilingual OSCAR text dataset.
format Preprint
id arxiv_https___arxiv_org_abs_2012_15834
institution arXiv
publishDate 2020
record_format arxiv
spellingShingle Loss Barcode: A Topological Measure of Escapability in Loss Landscapes
Barannikov, Serguei
Voronkova, Daria
Mironenko, Alexander
Trofimov, Ilya
Korotin, Alexander
Sotnikov, Grigorii
Burnaev, Evgeny
Machine Learning
Artificial Intelligence
Information Theory
Dynamical Systems
Neural network training is commonly based on SGD. However, the understanding of SGD's ability to converge to good local minima, given the non-convex nature of loss functions and the intricate geometric characteristics of loss landscapes, remains limited. In this paper, we apply topological data analysis methods to loss landscapes to gain insights into the learning process and generalization properties of deep neural networks. We use the loss function topology to relate the local behavior of gradient descent trajectories with the global properties of the loss surface. For this purpose, we define the neural network's Topological Obstructions score ("TO-score") with the help of robust topological invariants, barcodes of the loss function, which quantify the escapability of local minima for gradient-based optimization. Our two principal observations are: 1) the loss barcode of the neural network decreases with increasing depth and width, therefore the topological obstructions to learning diminish; 2) in certain situations there is a connection between the length of minima segments in the loss barcode and the minima's generalization errors. Our statements are based on extensive experiments with fully connected, convolutional, and transformer architectures and several datasets including MNIST, FMNIST, CIFAR10, CIFAR100, SVHN, and multilingual OSCAR text dataset.
title Loss Barcode: A Topological Measure of Escapability in Loss Landscapes
topic Machine Learning
Artificial Intelligence
Information Theory
Dynamical Systems
url https://arxiv.org/abs/2012.15834