Barcodes as Summary of Loss Function Topology

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Barannikov, Serguei, Korotin, Alexander, Oganesyan, Dmitry, Emtsev, Daniil, Burnaev, Evgeny
Format: Preprint
Publié: 2019
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866912184467980288
author Barannikov, Serguei
Korotin, Alexander
Oganesyan, Dmitry
Emtsev, Daniil
Burnaev, Evgeny
author_facet Barannikov, Serguei
Korotin, Alexander
Oganesyan, Dmitry
Emtsev, Daniil
Burnaev, Evgeny
contents We propose to study neural networks' loss surfaces by methods of topological data analysis. We suggest to apply barcodes of Morse complexes to explore topology of loss surfaces. An algorithm for calculations of the loss function's barcodes of local minima is described. We have conducted experiments for calculating barcodes of local minima for benchmark functions and for loss surfaces of small neural networks. Our experiments confirm our two principal observations for neural networks' loss surfaces. First, the barcodes of local minima are located in a small lower part of the range of values of neural networks' loss function. Secondly, increase of the neural network's depth and width lowers the barcodes of local minima. This has some natural implications for the neural network's learning and for its generalization properties.
format Preprint
id arxiv_https___arxiv_org_abs_1912_00043
institution arXiv
publishDate 2019
record_format arxiv
spellingShingle Barcodes as Summary of Loss Function Topology
Barannikov, Serguei
Korotin, Alexander
Oganesyan, Dmitry
Emtsev, Daniil
Burnaev, Evgeny
Machine Learning
Algebraic Topology
Dynamical Systems
Optimization and Control
We propose to study neural networks' loss surfaces by methods of topological data analysis. We suggest to apply barcodes of Morse complexes to explore topology of loss surfaces. An algorithm for calculations of the loss function's barcodes of local minima is described. We have conducted experiments for calculating barcodes of local minima for benchmark functions and for loss surfaces of small neural networks. Our experiments confirm our two principal observations for neural networks' loss surfaces. First, the barcodes of local minima are located in a small lower part of the range of values of neural networks' loss function. Secondly, increase of the neural network's depth and width lowers the barcodes of local minima. This has some natural implications for the neural network's learning and for its generalization properties.
title Barcodes as Summary of Loss Function Topology
topic Machine Learning
Algebraic Topology
Dynamical Systems
Optimization and Control
url https://arxiv.org/abs/1912.00043