Research map / Continual learning / Class-incremental
Persistent Backdoor Attacks in Class-Incremental Learning via Structural Invariant Anchoring
Junhuang Huang, Linshan Hou, Jianting Ning, Yanjun Zhang, Zhongyun Hua, Leo Yu Zhang
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 |
|---|---|---|---|
| add surrogate trajectory simulation to PBTO (ablation) | PBTO · CIFAR-100 after ten tasks | ASR (%): 62.1 -> 86.5 | Improved |
| add alignment loss Ls to PBTO (ablation) | PBTO · CIFAR-100 after ten tasks | ASR (%): 68.9 -> 86.5 | Improved |
| add trajectory simulation alone to plain trigger (ablation) | PBTO · CIFAR-100 after ten tasks | ASR (%): 4.5 -> 68.9 | Improved |
| increase surrogate trajectory length M from 1 to 5 (ablation) | PBTO · CIFAR-100 after ten tasks | ASR (%): 65.2 -> 86.5 | Improved |
| increase surrogate trajectory length M from 5 to 10 (ablation) | PBTO · CIFAR-100 after ten tasks | ASR (%): 86.5 -> 86.5 | No clear effect |
| increase proxy data from 1,000 to 5,000 images per class (ablation) | PBTO · CIFAR-100 after ten tasks | ASR (%): 76.8 -> 86.5 | Improved |
| increase proxy data from 5,000 to 10,000 images per class (ablation) | PBTO · CIFAR-100 after ten tasks | ASR (%): 86.5 -> 87.1 | No clear effect |
| use PBTO trigger instead of LTB under CIL | LTB · CIFAR-100 | ASR (%): 4.7 -> 86.5 | Improved |
| use PBTO instead of LTB on Tiny-ImageNet | LTB · Tiny-ImageNet | ASR (%): 8.9 -> 83.8 | Improved |
| transfer trigger from ResNet-18 to ViT-Small/16 | PBTO · CIFAR-100 | ASR (%): 86.5 -> 68.7 | Hurt |
| apply NAD defense after training | PBTO · CIFAR-10 | ASR (%): 97.1 -> 82.1 | Hurt |
| apply Fine-Pruning defense after training | PBTO · CIFAR-10 | ASR (%): 97.1 -> 88.4 | Hurt |
Closest papers by meaning
- Attack on Prompt: Backdoor Attack in Prompt-Based Continual Learning (AAAI 2025 · similarity 0.88)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.85)
- Clean Data Can Still Carry Backdoors: Support-Persistent Backdoors for Model Reuse (NeurIPS 2026 · similarity 0.85)
- ICLShield: Exploring and Mitigating In-Context Learning Backdoor Attacks (ICML 2025 · similarity 0.85)
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 2025 · similarity 0.84)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.84)
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.
