More is Not Always Better: Exploring Early Repair of DNNs

Andrei Mancu, Thomas Laurent, Franz Rieger, Paolo Arcaini, Fuyuki Ishikawa, Daniel Rueckert

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

3 Scopus citations

Abstract

DNN repair is an effective technique applied after training to enhance the class-specific accuracy of classifier models, where a low failure rate is required on specific classes. The repair methods introduced in recent studies assume that they are applied to fully trained models. In this paper, we argue that this could not always be the best choice. We analyse the performance of DNN models under various training times and repair combinations. Through meticulously designed experiments on two real-world datasets and a carefully curated assessment score, we show that applying DNN repair earlier in the training process, and not only at its end, can be beneficial. Thus, we encourage the research community to consider the idea of when to apply DNN repair in the model development.

Original languageEnglish
Title of host publicationProceedings - 2024 IEEE/ACM International Workshop on Deep Learning for Testing and Testing for Deep Learning, DeepTest 2024
PublisherAssociation for Computing Machinery, Inc
Pages13-16
Number of pages4
ISBN (Electronic)9798400705748
DOIs
StatePublished - 20 Apr 2024
Event2024 IEEE/ACM International Workshop on Deep Learning for Testing and Testing for Deep Learning, DeepTest 2024 - Lisbon, Portugal
Duration: 20 Apr 2024 → …

Publication series

NameProceedings - 2024 IEEE/ACM International Workshop on Deep Learning for Testing and Testing for Deep Learning, DeepTest 2024

Conference

Conference2024 IEEE/ACM International Workshop on Deep Learning for Testing and Testing for Deep Learning, DeepTest 2024
Country/TerritoryPortugal
CityLisbon
Period20/04/24 → …

Keywords

  • DNN repair
  • DNN training
  • deep neural networks
  • safety-critical

Fingerprint

Dive into the research topics of 'More is Not Always Better: Exploring Early Repair of DNNs'. Together they form a unique fingerprint.

Cite this