mlsanityData to 14 Oct 2026

InstantForget: Update-Free Backdoor Unlearning with Inference-Time Feature Reset

arXiv 2606.15730 ↗First posted 14 Jun 2026cs.LG cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Backdoor unlearning aims to remove a malicious trigger behavior from a deployed model while preserving clean utility. We study the update-free inference-time setting, where model parameters remain frozen. First, we audit a common projection assumption under oracle paired clean and triggered features. Projection succeeds mainly on BadNets and leaves WaNet, Blended, and SIG at 0.683, 0.888, and 0.941 ASR on CIFAR-10 ResNet-18. This failure is not explained by spectral compactness, spatial locality, or subspace misalignment. It is predicted by a logit-triplet gap involving the target margin, target-logit drop, and non-target logit rise. We then introduce InstantForget, a clean-calibrated gated reset that flags anomalous features with a Mahalanobis score and moves only flagged features toward a neutral non-target representation. With one fixed operating point selected on held-out triggered validation, InstantForget reduces average ASR to 0.071 across four non-adaptive CIFAR-10 triggers without triggered samples or parameter updates at deployment. It also reaches 0.981 detection AUROC and transfers to six of eight tested backbones. Reported failures under WaNet, ModelNet10 point blend, two backbone geometries, and adaptive feature-compactness attacks define the method's scope.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 33 of its 42 references that are in the map. First posted on arXiv 14 Jun 2026.

Builds on

33 of 42 references.

References · 33

  1. 2019Machine Unlearningcited by 1,061
  2. 2017Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoningcited by 1,055
  3. 2018Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networkscited by 580
  4. 2019Certified Data Removal from Machine Learning Modelscited by 535
  5. 2018Spectral Signatures in Backdoor Attackscited by 484
  6. 2019STRIP: A Defence Against Trojan Attacks on Deep Neural Networkscited by 459
  7. 2018Detecting Backdoor Attacks on Deep Neural Networks by Activation Clusteringcited by 456
  8. 2020Invisible Backdoor Attack with Sample-Specific Triggerscited by 353
  9. 2021Neural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networkscited by 324
  10. 2020Descent-to-Delete: Gradient-Based Methods for Machine Unlearningcited by 311
  11. 2023LEACE: Perfect linear concept erasure in closed formcited by 267
  12. 2021Anti-Backdoor Learning: Training Clean Models on Poisoned Datacited by 263
21 more
  1. 2019A new Backdoor Attack in CNNs by training set corruption without label poisoningcited by 242
  2. 2021Adversarial Neuron Pruning Purifies Backdoored Deep Modelscited by 235
  3. 2020Can AI help in screening Viral and COVID-19 pneumonia?cited by 200
  4. 2021Adversarial Unlearning of Backdoors via Implicit Hypergradientcited by 166
  5. 2022BackdoorBench: A Comprehensive Benchmark of Backdoor Learningcited by 128
  6. 2023SCALE-UP: An Efficient Black-box Input-level Backdoor Detection via Analyzing Scaled Prediction Consistencycited by 97
  7. 2023Reconstructive Neuron Pruning for Backdoor Defensecited by 59
  8. 2023Shared Adversarial Unlearning: Backdoor Mitigation by Unlearning Shared Adversarial Examplescited by 43
  9. 2024Machine Unlearning Fails to Remove Data Poisoning Attackscited by 37
  10. 2024Breaking the False Sense of Security in Backdoor Defense through Re-Activation Attackcited by 28
  11. 2025Certified Unlearning for Neural Networkscited by 27
  12. 2024The Utility and Complexity of in- and out-of-Distribution Machine Unlearningcited by 20
  13. 2024Targeted Unlearning with Single Layer Unlearning Gradientcited by 18
  14. 2022Backdoor Cleansing with Unlabeled Datacited by 17
  15. 2024Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy and Researchcited by 15
  16. 2025ForgetMe: Evaluating Selective Forgetting in Generative Modelscited by 11
  17. 2025Backdoor Token Unlearning: Exposing and Defending Backdoors in Pretrained Language Modelscited by 8
  18. 2026Class Unlearning via Depth-Aware Removal of Forget-Specific Directionscited by 4
  19. 2025ReVeil: Unconstrained Concealed Backdoor Attack on Deep Neural Networks using Machine Unlearningcited by 3
  20. 2025BURN: Backdoor Unlearning via Adversarial Boundary Analysiscited by 2
  21. 2025BadFU: Backdoor Federated Learning through Adversarial Machine Unlearningcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Kill it with FIRE: On Leveraging Latent Space Directions for Runtime Backdoor Mitigation in Deep Neural Networks
  2. 2025Revisiting Training-Inference Trigger Intensity in Backdoor Attacks
  3. 2026Removing the Trigger, Not the Backdoor: Alternative Triggers and Latent Backdoors
  4. 2025Improving LLM Unlearning Robustness via Random Perturbations
  5. 2025Isolate Trigger: Detecting and Eliminating Adaptive Backdoor Attacks ↗
  6. 2024Unlearn to Relearn Backdoors: Deferred Backdoor Functionality Attacks on Deep Learning Models
  7. 2026SpecGuard: Inference-Time Backdoor Detection For Free
  8. 2024Unlearning Backdoor Attacks through Gradient-Based Model Pruning