Research map / Adversarial robustness and security / Attacks on text and graphs
CoPur: Certifiably Robust Collaborative Inference via Feature Purification
Jing Liu, Chulin Xie, Oluwasanmi O Koyejo, Bo Li
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 CoPur feature purification instead of no defense | collaborative inference · NUS-WIDE targeted PGD | robust accuracy (%): 4.5±0.3 -> 83.7± 0.9 | Improved |
| use CoPur instead of Manifold Projection on ExtraSensory | collaborative inference · ExtraSensory targeted PGD | robust accuracy (%): 53.6 ± 35.7 -> 79.0± 5.0 | Improved |
| use Manifold Projection alone instead of no defense on NUS-WIDE | collaborative inference · NUS-WIDE targeted PGD | robust accuracy (%): 4.5±0.3 -> 81.3±1.2 | Improved |
| use CoPur instead of Manifold Projection on NUS-WIDE | collaborative inference · NUS-WIDE targeted PGD | robust accuracy (%): 81.3±1.2 -> 83.7± 0.9 | Improved |
| use adversarial training instead of no defense | collaborative inference · NUS-WIDE targeted PGD | robust accuracy (%): 4.5±0.3 -> 8.0±1.9 | Mixed |
| use block-wise adversarial training instead of standard adversarial training | adversarial training · ExtraSensory targeted PGD | robust accuracy (%): 19.8 ± 34.1 -> 0.0± 0.0 | Hurt |
| use CoPur under combined attack instead of Randomized Ablation | collaborative inference · ExtraSensory combined attack | robust accuracy (%): 60.3± 28.1 -> 79.1± 4.5 | Improved |
| increase feature-flipping amplification from 1 to 20 | baseline defenses · ExtraSensory untargeted | robust accuracy | Hurt |
| add more malicious agents (agents 7-10 vs 8-10) | CoPur · ExtraSensory untargeted | robust accuracy | Hurt |
| more than two malicious agents | Adversarial Training · ExtraSensory untargeted | robust accuracy | Hurt |
Closest papers by meaning
- Theoretical Insights in Model Inversion Robustness and Conditional Entropy Maximization for Collaborative Inference Systems (CVPR 2025 · similarity 0.86)
- Reimagining Mutual Information for Enhanced Defense against Data Leakage in Collaborative Inference (NeurIPS 2024 · similarity 0.84)
- CLIPure: Purification in Latent Space via CLIP for Adversarially Robust Zero-Shot Classification (ICLR 2025 · similarity 0.84)
- Adversarial Training on Purification (AToP): Advancing Both Robustness and Generalization (ICLR 2024 · similarity 0.83)
- Adversarial Purification with the Manifold Hypothesis (AAAI 2024 · similarity 0.83)
- Unextractable Protocol Models: Collaborative Training and Inference without Weight Materialization (NeurIPS 2025 · similarity 0.83)
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.
