Skip to main navigation Skip to search Skip to main content

Topological Insights into Sparse Neural Networks

  • S. Liu
  • , T. Van der Lee
  • , A. Yaman
  • , Z. Atashgahi
  • , D. Ferraro
  • , G. Sokar
  • , M. Pechenizkiy
  • , D.C. Mocanu

Research output: Chapter in Book / Report / Conference proceedingConference contributionAcademicpeer-review

Abstract

Sparse neural networks are effective approaches to reduce the resource requirements for the deployment of deep neural networks. Recently, the concept of adaptive sparse connectivity, has emerged to allow training sparse neural networks from scratch by optimizing the sparse structure during training. However, comparing different sparse topologies and determining how sparse topologies evolve during training, especially for the situation in which the sparse structure optimization is involved, remain as challenging open questions. This comparison becomes increasingly complex as the number of possible topological comparisons increases exponentially with the size of networks. In this work, we introduce an approach to understand and compare sparse neural network topologies from the perspective of graph theory. We first propose Neural Network Sparse Topology Distance (NNSTD) to measure the distance between different sparse neural networks. Further, we demonstrate that sparse neural networks can outperform over-parameterized models in terms of performance, even without any further structure optimization. To the end, we also show that adaptive sparse connectivity can always unveil a plenitude of sparse sub-networks with very different topologies which outperform the dense model, by quantifying and comparing their topological evolutionary processes. The latter findings complement the Lottery Ticket Hypothesis by showing that there is a much more efficient and robust way to find “winning tickets”. Altogether, our results start enabling a better theoretical understanding of sparse neural networks, and demonstrate the utility of using graph theory to analyze them.
Original languageEnglish
Title of host publicationMachine Learning and Knowledge Discovery in Databases
Subtitle of host publicationEuropean Conference, ECML PKDD 2020, Ghent, Belgium, September 14–18, 2020, Proceedings, Part III
EditorsFrank Hutter, Kristian Kersting, Jefrey Lijffijt, Isabel Valera
PublisherSpringer Science and Business Media Deutschland GmbH
Pages279-294
Number of pages16
Volume3
ISBN (Electronic)9783030676643
ISBN (Print)9783030676636
DOIs
Publication statusPublished - 2021
Externally publishedYes
EventEuropean Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2020 - Virtual, Online
Duration: 14 Sept 202018 Sept 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
PublisherSpringer
Volume12459 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349
NameECML PKDD: Joint European Conference on Machine Learning and Knowledge Discovery in Databases
PublisherSpringer
Volume2020

Conference

ConferenceEuropean Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2020
CityVirtual, Online
Period14/09/2018/09/20

Bibliographical note

© 2021, Springer Nature Switzerland AG.

Funding

This research has been partly funded by the NWO EDIC project. partly funded by the NWO EDIC

Funders
NWO EDIC
Nederlandse Organisatie voor Wetenschappelijk Onderzoek

    Fingerprint

    Dive into the research topics of 'Topological Insights into Sparse Neural Networks'. Together they form a unique fingerprint.

    Cite this