mlsanityData to 14 Oct 2026

Seal Your Backdoor with Variational Defense

arXiv 2503.08829 ↗First posted 11 Mar 2025cs.LG cs.CRcited by 2

Not on the map. Why?

Abstract (arXiv)

We propose VIBE, a model-agnostic framework that trains classifiers resilient to backdoor attacks. The key concept behind our approach is to treat malicious inputs and corrupted labels from the training dataset as observed random variables, while the actual clean labels are latent. VIBE then recovers the corresponding latent clean label posterior through variational inference. The resulting training procedure follows the expectation-maximization (EM) algorithm. The E-step infers the clean pseudolabels by solving an entropy-regularized optimal transport problem, while the M-step updates the classifier parameters via gradient descent. Being modular, VIBE can seamlessly integrate with recent advancements in self-supervised representation learning, which enhance its ability to resist backdoor attacks. We experimentally validate the method effectiveness against contemporary backdoor attacks on standard datasets, a large-scale setup with 1k classes, and a dataset poisoned with multiple attacks. VIBE consistently outperforms previous defenses across all tested scenarios.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 78 of its 116 references that are in the map. First posted on arXiv 11 Mar 2025.

Builds on

78 of 116 references.

References · 78

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  3. 2020Language Models are Few-Shot Learnerscited by 32,419
  4. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  5. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,364
  6. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  7. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  8. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,513
  9. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,403
  10. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
  11. 2013Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distancescited by 2,316
  12. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
38 more
  1. 2016Understanding deep learning requires rethinking generalizationcited by 1,674
  2. 2018Unsupervised Representation Learning by Predicting Image Rotationscited by 1,591
  3. 2016Context Encoders: Feature Learning by Inpaintingcited by 1,428
  4. 2016Unsupervised Learning of Visual Representations by Solving Jigsaw Puzzlescited by 1,382
  5. 2022Reproducible scaling laws for contrastive language-image learningcited by 1,173
  6. 2017Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoningcited by 1,055
  7. 2018Computational Optimal Transportcited by 780
  8. 2019Big Transfer (BiT): General Visual Representation Learningcited by 769
  9. 2018Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networkscited by 580
  10. 2019Self-labelling via simultaneous clustering and representation learningcited by 528
  11. 2018Poison Frogs! Targeted Clean-Label Poisoning Attacks on Neural Networkscited by 519
  12. 2018Spectral Signatures in Backdoor Attackscited by 484
  13. 2018Detecting Backdoor Attacks on Deep Neural Networks by Activation Clusteringcited by 456
  14. 2018From Louvain to Leiden: guaranteeing well-connected communitiescited by 385
  15. 2020Invisible Backdoor Attack with Sample-Specific Triggerscited by 353
  16. 2020The Pitfalls of Simplicity Bias in Neural Networkscited by 351
  17. 2021Neural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networkscited by 324
  18. 2019Label-Consistent Backdoor Attackscited by 296
  19. 2021Anti-Backdoor Learning: Training Clean Models on Poisoned Datacited by 263
  20. 2019A new Backdoor Attack in CNNs by training set corruption without label poisoningcited by 242
  21. 2021Adversarial Neuron Pruning Purifies Backdoored Deep Modelscited by 235
  22. 2019SGD on Neural Networks Learns Functions of Increasing Complexitycited by 204
  23. 2021Adversarial Unlearning of Backdoors via Implicit Hypergradientcited by 166
  24. 2020Witches' Brew: Industrial Scale Data Poisoning via Gradient Matchingcited by 163
  25. 2022Backdoor Defense via Decoupling the Training Processcited by 153
  26. 2020Predicting What You Already Know Helps: Provable Self-Supervised Learningcited by 152
  27. 2022On Feature Learning in the Presence of Spurious Correlationscited by 145
  28. 2014Statistical guarantees for the EM algorithm: From population to sample-based analysiscited by 133
  29. 2020Bridging Mode Connectivity in Loss Landscapes and Adversarial Robustnesscited by 132
  30. 2023BadCLIP: Dual-Embedding Guided Backdoor Attack on Multimodal Contrastive Learningcited by 115
  31. 2019TBT: Targeted Neural Network Attack with Bit Trojancited by 113
  32. 2021Backdoor Scanning for Deep Neural Networks through K-Arm Optimizationcited by 98
  33. 2023SCALE-UP: An Efficient Black-box Input-level Backdoor Detection via Analyzing Scaled Prediction Consistencycited by 97
  34. 2020Deep Feature Space Trojan Attack of Neural Networks by Controlled Detoxificationcited by 95
  35. 2021Backdoor Attacks on Self-Supervised Learningcited by 90
  36. 2019Defending Neural Backdoors via Generative Distribution Modelingcited by 84
  37. 2021Handcrafted Backdoors in Deep Neural Networkscited by 63
  38. 2023Reconstructive Neuron Pruning for Backdoor Defensecited by 59

50 most cited of 78 All 78 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026From Internal Diagnosis to External Auditing: A VLM-Driven Paradigm for Data-Free Online Backdoor Defensecited by 3
  2. 2026BYORn: Bootstrap Your Own Responses to Defend Large Vision-Language Models Against Backdoor Attackscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Prototype-Guided Robust Learning against Backdoor Attacks
  2. 2022Under-confidence Backdoors Are Resilient and Stealthy Backdoors
  3. 2024Gracefully Filtering Backdoor Samples for Generative Large Language Models without Retraining
  4. 2022Planting Undetectable Backdoors in Machine Learning Models
  5. 2025The Eminence in Shadow: Exploiting Feature Boundary Ambiguity for Robust Backdoor Attacks
  6. 2024SEEP: Training Dynamics Grounds Latent Representation Search for Mitigating Backdoor Poisoning Attacks
  7. 2026Benign on Label, Malicious by Design: Clean-Label Dormant-to-Activated Backdoor via Machine Unlearning with Removable Camouflage ↗
  8. 2022BackdoorBench: A Comprehensive Benchmark of Backdoor Learning