Research map / Adversarial robustness and security / Backdoor and poisoning
Defense against backdoor attacks via robust covariance estimation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace squared norm score (alpha=0) with QUE score alpha=4 (ablation) | SPECTRE · 1-way eps=0.1 | remaining poisoned samples: 3 -> 0 | Hurt |
| replace squared norm score (alpha=0) with QUE score alpha=4 (ablation) | SPECTRE · 2-way eps=0.05 | remaining poisoned samples: 69 -> 30 | Helped |
| replace squared norm score (alpha=0) with QUE score alpha=4 (ablation) | SPECTRE · 3-way eps=0.0124 | remaining poisoned samples: 22 -> 5 | Helped |
| replace projected norm (alpha=inf) with QUE alpha=4 (ablation) | SPECTRE · 1-way eps=0.1 | remaining poisoned samples: 118 -> 0 | Helped |
| replace projected norm (alpha=inf) with QUE alpha=4 (ablation) | SPECTRE · 2-way eps=0.05 | remaining poisoned samples: 97 -> 30 | Helped |
| replace projected norm (alpha=inf) with QUE alpha=4 (ablation) | SPECTRE · 3-way eps=0.0124 | remaining poisoned samples: 5 -> 5 | No clear effect |
| use fixed k=20 instead of adaptive Algorithm 3 (ablation) | SPECTRE · poison removal across attacks | mean prm/(eps*n) (%): 98.2 -> 76.5 | Hurt |
| use fixed k=100 instead of adaptive Algorithm 3 (ablation) | SPECTRE · poison removal across attacks | mean prm/(eps*n) (%): 98.2 -> 86.8 | Hurt |
| use fixed k=20 instead of adaptive Algorithm 3 (ablation) | SPECTRE · poison removal across attacks | min prm/(eps*n) (%): 87.1 -> 0.0 | Hurt |
| use fixed k=100 instead of adaptive Algorithm 3 (ablation) | SPECTRE · poison removal across attacks | min prm/(eps*n) (%): 87.1 -> 4.0 | Hurt |
| replace PCA defense with SPECTRE (ablation) | retrained ResNet-32 · CIFAR-10 2-way pixel attack, epsilon*n=125 | backdoor attack accuracy: 0.139 -> 0.000 | Helped |
| replace PCA defense with SPECTRE (ablation) | retrained ResNet-32 · CIFAR-10 1-way pixel attack, epsilon*n=250 | backdoor attack accuracy: 0.880 -> 0.001 | Helped |
Closest papers by meaning
- Physical Backdoor: Towards Temperature-based Backdoor Attacks in the Physical World (CVPR 2024 · similarity 0.85)
- Provable Robustness against Backdoor Attacks via the Primal-Dual Perspective on Differential Privacy (NeurIPS 2026 · similarity 0.84)
- RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance (NeurIPS 2026 · similarity 0.83)
- Backdoor Secrets Unveiled: Identifying Backdoor Data with Optimized Scaled Prediction Consistency (ICLR 2024 · similarity 0.82)
- Unelicitable Backdoors via Cryptographic Transformer Circuits (NeurIPS 2024 · similarity 0.81)
- Private Adaptive Covariance Estimation via Gaussian Graphical Models (NeurIPS 2026 · similarity 0.81)
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.