mlsanityData to 7 Oct 2026

Low-Loss Subspace Compression for Clean Gains against Multi-Agent Backdoor Attacks

arXiv 2203.03692 ↗First posted 7 Mar 2022cs.LG cs.CR cs.MAcited by 4

On the map: LLM safety and security → Prompt injection and agent security

Abstract (arXiv)

Recent exploration of the multi-agent backdoor attack demonstrated the backfiring effect, a natural defense against backdoor attacks where backdoored inputs are randomly classified. This yields a side-effect of low accuracy w.r.t. clean labels, which motivates this paper's work on the construction of multi-agent backdoor defenses that maximize accuracy w.r.t. clean labels and minimize that of poison labels. Founded upon agent dynamics and low-loss subspace construction, we contribute three defenses that yield improved multi-agent backdoor robustness.

Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Placed in Prompt injection and agent security, one of 12 directions in LLM safety and security. Builds on 41 of its 51 references that are in the map. First posted on arXiv 7 Mar 2022.

Builds on

41 of 51 references.

Closest ancestors · 1

Same direction, newest first

  1. 2022Backdoors Stuck At The Frontdoor: Multi-Agent Backdoor Attacks That Backfire28 Jan

Other references · 40

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
  3. 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,888
  4. 2019CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Featurescited by 2,875
  5. 2017Arbitrary Style Transfer in Real-time with Adaptive Instance Normalizationcited by 1,998
  6. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,430
  7. 2018Averaging Weights Leads to Wider Optima and Better Generalizationcited by 1,194
  8. 2017BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chaincited by 1,052
  9. 2017Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoningcited by 1,012
  10. 2018How To Backdoor Federated Learningcited by 745
  11. 2019Linear Mode Connectivity and the Lottery Ticket Hypothesiscited by 627
  12. 2018Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNscited by 585
28 more
  1. 2018Poison Frogs! Targeted Clean-Label Poisoning Attacks on Neural Networkscited by 500
  2. 2019Local Model Poisoning Attacks to Byzantine-Robust Federated Learningcited by 479
  3. 2018Spectral Signatures in Backdoor Attackscited by 457
  4. 2019Deep Ensembles: A Loss Landscape Perspectivecited by 438
  5. 2018Detecting Backdoor Attacks on Deep Neural Networks by Activation Clusteringcited by 429
  6. 2020Backdoor Learning: A Surveycited by 412
  7. 2018Measuring the Intrinsic Dimension of Objective Landscapescited by 364
  8. 2019Hidden Trigger Backdoor Attackscited by 359
  9. 2018Essentially No Barriers in Neural Network Energy Landscapecited by 328
  10. 2020Attack of the Tails: Yes, You Really Can Backdoor Federated Learningcited by 307
  11. 2019Can You Really Backdoor Federated Learning?cited by 250
  12. 2020Supermasks in Superpositioncited by 223
  13. 2014Qualitatively characterizing neural network optimization problemscited by 186
  14. 2020Linear Mode Connectivity in Multitask and Continual Learningcited by 144
  15. 2019Transferable Clean-Label Poisoning Attacks on Deep Neural Netscited by 129
  16. 2020Backdoor Attacks and Countermeasures on Deep Learning: A Comprehensive Reviewcited by 120
  17. 2019Subspace Inference for Bayesian Deep Learningcited by 94
  18. 2020Strong Data Augmentation Sanitizes Poisoning and Backdoor Attacks Without an Accuracy Tradeoffcited by 86
  19. 2021Learning Neural Network Subspacescited by 82
  20. 2019Large Scale Structure of Neural Network Loss Landscapescited by 68
  21. 2019HyperGAN: A Generative Model for Diverse, Performant Neural Networkscited by 51
  22. 2020Neural networks with late-phase weightscited by 29
  23. 2018Universal Multi-Party Poisoning Attackscited by 17
  24. 2019Contamination Attacks and Mitigation in Multi-Party Machine Learningcited by 13
  25. 2020Dynamic backdoor attacks against federated learningcited by 9
  26. 2019Biometric Backdoors: A Poisoning Attack Against Unsupervised Template Updatingcited by 7
  27. 2021LCS: Learning Compressible Subspaces for Adaptive Network Compression at Inference Timecited by 3
  28. 2022Hiding Behind Backdoors: Self-Obfuscation Against Generative Modelscited by 2

Built on by

Cited by year

Citing papers per year: 2022: 3; 2023: 1.

4 papers.

  1. 2022Learn2Weight: Parameter Adaptation against Similar-domain Adversarial Attackscited by 5
  2. 2022GreaseVision: Rewriting the Rules of the Interfacecited by 2
  3. 2022Cross-Reality Re-Rendering: Manipulating between Digital and Physical Realitiescited by 0
  4. 2023Projected Subnetworks Scale Adaptationcited by 0