mlsanityData to 14 Oct 2026

Prototype Guided Backdoor Defense

arXiv 2503.20925 ↗First posted 26 Mar 2025cs.CV cs.AI cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Deep learning models are susceptible to em backdoor attacks involving malicious attackers perturbing a small subset of training data with a em trigger to causes misclassifications. Various triggers have been used, including semantic triggers that are easily realizable without requiring the attacker to manipulate the image. The emergence of generative AI has eased the generation of varied poisoned samples. Robustness across types of triggers is crucial to effective defense. We propose Prototype Guided Backdoor Defense (PGBD), a robust post-hoc defense that scales across different trigger types, including previously unsolved semantic triggers. PGBD exploits displacements in the geometric spaces of activations to penalize movements toward the trigger. This is done using a novel sanitization loss of a post-hoc fine-tuning step. The geometric approach scales easily to all types of attacks. PGBD achieves better performance across all settings. We also present the first defense against a new semantic attack on celebrity face images. Project page: hyperlinkhttps://venkatadithya9.github.io/pgbd.github.io/this https URL.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 55 of its 78 references that are in the map. First posted on arXiv 26 Mar 2025.

Builds on

55 of 78 references.

References · 55

  1. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,968
  2. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  3. 2016Identity Mappings in Deep Residual Networkscited by 1,965
  4. 2019Certified Adversarial Robustness via Randomized Smoothingcited by 1,340
  5. 2017Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoningcited by 1,055
  6. 2020Concept Bottleneck Modelscited by 1,032
  7. 2018Spectral Signatures in Backdoor Attackscited by 484
  8. 2019STRIP: A Defence Against Trojan Attacks on Deep Neural Networkscited by 459
  9. 2018Detecting Backdoor Attacks on Deep Neural Networks by Activation Clusteringcited by 456
  10. 2020Invisible Backdoor Attack with Sample-Specific Triggerscited by 353
  11. 2021Neural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networkscited by 324
  12. 2020Input-Aware Dynamic Backdoor Attackcited by 313
38 more
  1. 2021Anti-Backdoor Learning: Training Clean Models on Poisoned Datacited by 263
  2. 2019A new Backdoor Attack in CNNs by training set corruption without label poisoningcited by 242
  3. 2021Adversarial Neuron Pruning Purifies Backdoored Deep Modelscited by 235
  4. 2019Detecting AI Trojans Using Meta Neural Analysiscited by 199
  5. 2022Neural Feature Fusion Fields: 3D Distillation of Self-Supervised 2D Image Representationscited by 196
  6. 2020Dataset Security for Machine Learning: Data Poisoning, Backdoor Attacks, and Defensescited by 180
  7. 2021Adversarial Unlearning of Backdoors via Implicit Hypergradientcited by 166
  8. 2018SentiNet: Detecting Localized Universal Attacks Against Deep Learning Systemscited by 152
  9. 2020Dynamic Backdoor Attacks Against Machine Learning Modelscited by 151
  10. 2019Universal Litmus Patterns: Revealing Backdoor Attacks in CNNscited by 146
  11. 2020Backdoor Attacks Against Deep Learning Systems in the Physical Worldcited by 134
  12. 2021SPECTRE: Defending Against Backdoor Attacks Using Robust Statisticscited by 130
  13. 2022BackdoorBench: A Comprehensive Benchmark of Backdoor Learningcited by 128
  14. 2021Sleeper Agent: Scalable Hidden Trigger Backdoors for Neural Networks Trained from Scratchcited by 102
  15. 2020An Embarrassingly Simple Approach for Trojan Attack in Deep Neural Networkscited by 97
  16. 2020Deep Feature Space Trojan Attack of Neural Networks by Controlled Detoxificationcited by 95
  17. 2020Intrinsic Certified Robustness of Bagging against Data Poisoning Attackscited by 94
  18. 2022Wild Patterns Reloaded: A Survey of Machine Learning Security against Training Data Poisoningcited by 94
  19. 2021Backdoor Attacks on Self-Supervised Learningcited by 90
  20. 2020Strong Data Augmentation Sanitizes Poisoning and Backdoor Attacks Without an Accuracy Tradeoffcited by 89
  21. 2022Interactive Concept Bottleneck Modelscited by 79
  22. 2018Label Sanitization against Label Flipping Poisoning Attackscited by 73
  23. 2021BACKDOORL: Backdoor Attack against Competitive Reinforcement Learningcited by 68
  24. 2022Backdoor Defense with Machine Unlearningcited by 65
  25. 2023Enhancing Fine-Tuning Based Backdoor Defense with Sharpness-Aware Minimizationcited by 64
  26. 2023Reconstructive Neuron Pruning for Backdoor Defensecited by 59
  27. 2023Detecting Backdoors in Pre-trained Encoderscited by 55
  28. 2022Improved Certified Defenses against Data Poisoning with (Deterministic) Finite Aggregationcited by 52
  29. 2022Concept-level Debugging of Part-Prototype Networkscited by 52
  30. 2021Trigger Hunting with a Topological Prior for Trojan Detectioncited by 39
  31. 2023Towards Stable Backdoor Purification through Feature Shift Tuningcited by 32
  32. 2022An Embarrassingly Simple Backdoor Attack on Self-supervised Learningcited by 31
  33. 2022Proto2Proto: Can you recognize the car, the way I do?cited by 23
  34. 2019On Defending Against Label Flipping Attacks on Malware Detection Systemscited by 22
  35. 2023Defending Against Patch-based Backdoor Attacks on Self-Supervised Learningcited by 20
  36. 2021DeepObliviate: A Powerful Charm for Erasing Data Residual Memory in Deep Neural Networkscited by 19
  37. 2021On Recognizing Occluded Faces in the Wildcited by 15
  38. 2021Enhancing Backdoor Attacks with Multi-Level MMD Regularizationcited by 15

50 most cited of 55 All 55 (JSON)

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Proactive Adversarial Defense: Harnessing Prompt Tuning in Vision-Language Models to Detect Unseen Backdoored Images
  2. 2021Rethinking the Backdoor Attacks' Triggers: A Frequency Perspective
  3. 2022M-to-N Backdoor Paradigm: A Multi-Trigger and Multi-Target Attack to Deep Learning Models
  4. 2023SoK: A Systematic Evaluation of Backdoor Trigger Characteristics in Image Classification
  5. 2022Imperceptible Backdoor Attack: From Input Space to Feature Representation
  6. 2022Architectural Backdoors in Neural Networks
  7. 2025Prototype-Guided Robust Learning against Backdoor Attacks
  8. 2024Attack as Defense: Run-time Backdoor Implantation for Image Content Protection ↗