Fine-pruning: Defending against backdooring attacks on deep neural networks

Kang Liu, Brendan Dolan-Gavitt, Siddharth Garg

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Deep neural networks (DNNs) provide excellent performance across a wide range of classification tasks, but their training requires high computational resources and is often outsourced to third parties. Recent work has shown that outsourced training introduces the risk that a malicious trainer will return a backdoored DNN that behaves normally on most inputs but causes targeted misclassifications or degrades the accuracy of the network when a trigger known only to the attacker is present. In this paper, we provide the first effective defenses against backdoor attacks on DNNs. We implement three backdoor attacks from prior work and use them to investigate two promising defenses, pruning and fine-tuning. We show that neither, by itself, is sufficient to defend against sophisticated attackers. We then evaluate fine-pruning, a combination of pruning and fine-tuning, and show that it successfully weakens or even eliminates the backdoors, i.e., in some cases reducing the attack success rate to 0% with only a 0.4% drop in accuracy for clean (non-triggering) inputs. Our work provides the first step toward defenses against backdoor attacks in deep neural networks.

Original languageEnglish (US)
Title of host publicationResearch in Attacks, Intrusions, and Defenses - 21st International Symposium, RAID 2018, Proceedings
EditorsMichael Bailey, Sotiris Ioannidis, Manolis Stamatogiannakis, Thorsten Holz
PublisherSpringer Verlag
Pages273-294
Number of pages22
ISBN (Print)9783030004699
DOIs
StatePublished - 2018
Event21st International Symposium on Research in Attacks, Intrusions and Defenses, RAID 2018 - Heraklion, Greece
Duration: Sep 10 2018Sep 12 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11050 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other21st International Symposium on Research in Attacks, Intrusions and Defenses, RAID 2018
Country/TerritoryGreece
CityHeraklion
Period9/10/189/12/18

Keywords

  • Backdoor
  • Deep learning
  • Fine-tuning
  • Pruning
  • Trojan

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Fine-pruning: Defending against backdooring attacks on deep neural networks'. Together they form a unique fingerprint.

Cite this