Research map / Label-efficient and robust learning / Recognition and long-tail
From Measurement to Mitigation: Quantifying and Reducing Identity Leakage in Image Representation Encoders with Linear Subspace Removal
Daniel George, Charles Yeh, Daniel Lee, Yifei 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 |
|---|---|---|---|
| Apply ISP projection | CLIP · CelebA-20 (k=16) | Ridge open-set TAR@FAR=1e-4 (%): 19.8% -> 13.0% | Improved |
| Apply ISP projection | CLIP · VGGFace2-20 (k=16) | Ridge open-set TAR@FAR=1e-4 (%): 14.1% -> 2.3% | Improved |
| Apply ISP projection | DINOv3 · CelebA-20 (k=16) | Ridge open-set TAR@FAR=1e-4 (%): 6.8% -> 2.8% | Improved |
| Apply ISP projection | SSCD · VGGFace2-20 (k=16) | Ridge open-set TAR@FAR=1e-4 (%): 4.5% -> 0.5% | Improved |
| Apply ISP projection | DINOv2 · VGGFace2-20 (k=16) | Ridge open-set TAR@FAR=1e-4 (%): 1.6% -> 1.5% | No clear effect |
| Fit ISP on the other dataset (cross-dataset transfer) | CLIP · CelebA-20 (k=16) | Ridge open-set TAR@FAR=1e-4 (%): 19.8% -> 10.2% | Improved |
| Fit ISP on the other dataset (cross-dataset transfer) | DINOv2 · VGGFace2-20 (k=16) | Ridge open-set TAR@FAR=1e-4 (%): 1.6% -> 2.7% | Mixed |
| Apply ISP projection | DINOv2, DINOv3, CLIP · ImageNet k-NN | Top-1 accuracy, normalized: 100.1%, 97.3%, 98.3% of raw | No clear effect |
| Apply ISP projection | SSCD · ImageNet k-NN | Top-1 accuracy, normalized: 85.4% of raw | Hurt |
| Use ISP instead of LEACE | DINOv3 · ImageNet k-NN | Top-1 accuracy, normalized: 97.3% vs 97.4% | No clear effect |
| Apply ISP projection | Non-FR encoders · Template inversion | Verification rate | No clear effect |
| Apply ISP projection | Non-FR encoders · Non-linear MLP probe | TAR | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LEACE — LEACE: Perfect linear concept erasure in closed form (NeurIPS 2023) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.86)
- Privacy-Preserving Face Recognition Using Trainable Feature Subtraction (CVPR 2024 · similarity 0.85)
- How Private is Private? A Comparative Study for Face De-Identification (NeurIPS 2026 · similarity 0.85)
- Transferable Facial Privacy Protection against Blind Face Restoration via Domain-Consistent Adversarial Obfuscation (ICML 2024 · similarity 0.85)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.84)
- FSFM: A Generalizable Face Security Foundation Model via Self-Supervised Facial Representation Learning (CVPR 2025 · 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.
