Research map / Adversarial robustness and security / Backdoor and poisoning
Breaking the False Sense of Security in Backdoor Defense through Re-Activation Attack
Mingli Zhu, Siyuan Liang, Baoyuan Wu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Re-investigates backdoored models after defenses and shows that supposedly eliminated backdoors can be re-activated by a small amount of fine-tuning, with a re-activation attack exposing the false sense of security in backdoor defenses.
Problem
Existing backdoor defenses reduce attack success rates, but it is unclear whether the backdoor is truly eliminated from the model; the paper asks whether defended models remain vulnerable to reactivation of the backdoor.
Approach
The paper shows that backdoor defenses mainly suppress rather than remove backdoor features, so a backdoor can be re-activated quickly by fine-tuning the defended model on a small set of poisoned-looking data, and it proposes a re-activation attack framework to evaluate defenses accordingly.
Surprising findings
- Defended models that show low attack success rates can have their backdoor re-activated quickly by fine-tuning on a small amount of data, suggesting defenses suppress rather than remove backdoors.
Limitations
- Provided text is only the NeurIPS template; the actual experimental details are not available to verify results.
Experimental setup
Metrics: attack success rate
Baselines: fine-pruning neural attention distillation anti-backdoor learning i-BAU backdoor-safe fine-tuning
Builds on: BadNets Blended attack
Closest papers by meaning
- REFINE: Inversion-Free Backdoor Defense via Model Reprogramming (ICLR 2025 · similarity 0.92)
- Progressive Backdoor Erasing via Connecting Backdoor and Adversarial Attacks (CVPR 2023 · similarity 0.92)
- BAN: Detecting Backdoors Activated by Adversarial Neuron Noise (NeurIPS 2024 · similarity 0.91)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.90)
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · similarity 0.90)
- Uncovering, Explaining, and Mitigating the Superficial Safety of Backdoor Defense (NeurIPS 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
