Research map / Adversarial robustness and security / Backdoor and poisoning
Theory of Minimal Weight Perturbations in Deep Networks and its Applications for Low-Rank Activated Backdoor Attacks
Bethan Evans, Jared Tanner
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase regularisation lambda to 200 when training the perturbation (ablation) | Empirical single-layer weight perturbation · 5-layer network, 4-class synthetic | class flip success: flips at lambda=1 in all layers; none at lambda=200 | Hurt |
| perturb layers 1 to k jointly instead of a single layer (ablation) | 10-layer linear network · 4-class synthetic | perturbation norm needed | Improved |
| apply low-rank projection of the final layer to a backdoor trained with the precision-conditioned loss | VGG16 on CIFAR-10 · CIFAR-10 rank 10 to rank 8 | ASR: 10.1 -> 96.2 | Improved |
| use precision-conditioned loss instead of control loss (ablation) | VGG16 on CIFAR-10 · CIFAR-10 full precision | ASR at full precision: 97.0 -> 10.1 | Improved |
| apply low-rank compression to Phi-2 fc2 layers | Phi-2 backdoor · SQuAD | ASR: 45.5 -> 86.0 | Improved |
| prune the backdoored ResNet18 to higher sparsity (ablation) | ResNet18 on CIFAR-10 · single sample | margin bound satisfied / class flip: class flip first at 17% sparsity | Improved |
Closest papers by meaning
- Towards Certificated Model Robustness Against Weight Perturbations (AAAI 2020 · similarity 0.90)
- Lipschitz-Margin Training: Scalable Certification of Perturbation Invariance for Deep Neural Networks (NeurIPS 2018 · similarity 0.86)
- Fooling a Complete Neural Network Verifier (ICLR 2021 · similarity 0.86)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.85)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.85)
- Backdoor Channels Hidden in Latent Space: Cryptographic Undetectability in Modern Neural Networks (NeurIPS 2026 · similarity 0.85)
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.
