Research map / Adversarial robustness and security / Backdoor and poisoning
Trigger Hunting with a Topological Prior for Trojan Detection
Xiaoling Hu, Xiao Lin, Michael Cogswell, Yi Yao, Susmit Jha, Chao Chen
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 |
|---|---|---|---|
| Use topological and diversity losses with three trigger candidates | Trojan detector · Trojaned-MNIST and Trojaned-CIFAR10 | AUC: 0.88 $\pm$ 0.04 vs 0.65 $\pm$ 0.07 (MNIST); 0.91 $\pm$ 0.05 vs 0.75 $\pm$ 0.07 (CIFAR10) | Improved |
| Use topological and diversity losses with three trigger candidates | Trojan detector · TrojAI-Round1 to Round4 | AUC: 0.90 $\pm$ 0.02 vs 0.71 $\pm$ 0.04; 0.87 $\pm$ 0.05 vs 0.66 $\pm$ 0.07; 0.89 $\pm$ 0.04 vs 0.62 $\pm$ 0.07; 0.92 $\pm$ 0 | Improved |
| Add the topological loss (ablation) | Trigger search and Trojan detector · TrojAI-Round4 | AUC: 0.89 $\pm$ 0.04 -> 0.92 $\pm$ 0.06 | Improved |
| Add the diversity loss and generate multiple candidates (N_T=3) (ablation) | Trigger search and Trojan detector · TrojAI-Round4 | AUC: 0.85 $\pm$ 0.02 -> 0.92 $\pm$ 0.06 | Improved |
| Use three rather than two trigger candidates (ablation) | Trigger search and Trojan detector · TrojAI-Round4 | AUC: 0.89 $\pm$ 0.05 -> 0.92 $\pm$ 0.06 | Improved |
| Add the topological loss (ablation) | Trigger search and Trojan detector with 25 training samples · TrojAI-Round4 | AUC: 0.73 $\pm$ 0.03 -> 0.77 $\pm$ 0.04 | Improved |
| Add the diversity loss (ablation) | Trigger search and Trojan detector with 25 training samples · TrojAI-Round4 | AUC: 0.68 $\pm$ 0.04 -> 0.77 $\pm$ 0.04 | Improved |
| Use the topological loss (ablation) | Trigger search · Trigger reconstruction | iterations to converge: ≈50 -> ≈30 | Improved |
| Vary the topological-loss weight (ablation) | Trigger search and Trojan detector · TrojAI-Round4 | AUC | No clear effect |
Closest papers by meaning
- Rethinking the Reverse-engineering of Trojan Triggers (NeurIPS 2022 · similarity 0.93)
- CatchBackdoor: Backdoor Detection via Critical Trojan Neural Path Fuzzing (ECCV 2024 · similarity 0.91)
- Deep-TROJ: An Inference Stage Trojan Insertion Algorithm through Efficient Weight Replacement Attack (CVPR 2024 · similarity 0.86)
- TrojanDec: Data-free Detection of Trojan Inputs in Self-supervised Learning (AAAI 2025 · similarity 0.86)
- Scanning Trojaned Models Using Out-of-Distribution Samples (NeurIPS 2024 · similarity 0.85)
- Trapping Attacker in Dilemma: Examining Internal Correlations and External Influences of Trigger for Defending GNN Backdoors (NeurIPS 2026 · 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.
