Research map / Federated learning and privacy / Privacy attacks and secure learning
Federated Learning for Face Recognition with Gradient Correction
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 |
|---|---|---|---|
| Add softmax gradient correction to FedPE | Softmax-FedPE with ResNet-34 · LFW, CFP-FP, AgeDB, CALFW, CPLFW, SLLFW, VGG2-FP | verification accuracy (%): 94.77 -> 98.40; 81.90 -> 90.20; 78.38 -> 85.85; 74.15 -> 81.47; 64.40 -> 71.88; 80.42 -> 90.38; 80.32 -> 87.64 | Helped |
| Add softmax gradient correction to FedPE | Softmax-FedPE with ResNet-34 · MegaFace | verification TPR (@FPR=1e-6) (%): 36.83 -> 69.87 | Helped |
| Add softmax gradient correction to FedPE | Softmax-FedPE with ResNet-34 · MegaFace | identification Rank-1 (%): 34.08 -> 61.26 | Helped |
| Add softmax gradient correction to FedPE | Softmax-FedPE with ResNet-34 · IJB-B and IJB-C | verification TPR (@FPR=1e-3) (%): 54.33 -> 69.23; 57.85 -> 71.33 | Helped |
| Add softmax gradient correction to FedPE | Softmax-FedPE with ResNet-34 · IJB-B and IJB-C | identification Rank-1 (%): 64.44 -> 78.52; 65.35 -> 79.52 | Helped |
| Set regularizer multiplier λ to 20 rather than 1 or 50 (ablation) | Softmax-FedGC · LFW | verification accuracy (%): 95.20; 98.40; 97.42 | Helped |
| Replace the softmax regularizer with a cosine regularizer (ablation) | FedGC · LFW | verification accuracy (%): 98.40 -> 96.63 | Hurt |
| Increase the participant fraction from C=0.25 to C=1 (ablation) | Softmax-FedGC · LFW | verification performance: 97.07 -> 98.40 | Helped |
| Use an unbalanced rather than balanced data partition (ablation) | FedGC · LFW, CFP-FP, AgeDB | verification accuracy (%): 98.40 -> 98.80; 90.20 -> 91.56; 85.85 -> 88.78 | Helped |
| Keep the final fully-connected layer fixed rather than adaptively optimizing it (ablation) | FedGC · Softmax, CosFace and ArcFace verification benchmarks | average accuracy: 88.32 -> 81.60; 90.86 -> 79.18; 85.06 -> 75.42 | Hurt |
Closest papers by meaning
- Federated Learning via Input-Output Collaborative Distillation (AAAI 2024 · similarity 0.86)
- Little Is Enough: Boosting Privacy by Sharing Only Hard Labels in Federated Semi-Supervised Learning (AAAI 2025 · similarity 0.86)
- (FL)$^2$: Overcoming Few Labels in Federated Semi-Supervised Learning (NeurIPS 2024 · similarity 0.86)
- Clients Collaborate: Flexible Differentially Private Federated Learning with Guaranteed Improvement of Utility-Privacy Trade-off (ICML 2025 · similarity 0.85)
- FedSC: Provable Federated Self-supervised Learning with Spectral Contrastive Objective over Non-i.i.d. Data (ICML 2024 · similarity 0.85)
- FedMSGL: A Self-Expressive Hypergraph Based Federated Multi-View Learning (AAAI 2025 · similarity 0.85)
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.