Research map / Label-efficient and robust learning / Recognition and long-tail
Loss Function Search for Face Recognition
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 |
|---|---|---|---|
| Reduce softmax probability using a negative modulating factor instead of using the unmodified probability (a=0) (ablation) | Random-Softmax · LFW, MS-Celeb-1M-v1c-R | accuracy: 99.53 -> 99.56 (a=-1); 99.53 -> 99.66 (a=-10); 99.53 -> 99.71 (a=-100) | Helped |
| Reduce softmax probability using a negative modulating factor instead of using the unmodified probability (a=0) (ablation) | Random-Softmax · SLLFW, MS-Celeb-1M-v1c-R | accuracy: 98.78 -> 98.91 (a=-1); 98.78 -> 99.20 (a=-10); 98.78 -> 99.28 (a=-100) | Helped |
| Increase sampled candidate models from B=2 to B=4 (ablation) | Search-Softmax · SLLFW, MS-Celeb-1M-v1c-R | accuracy: 99.31 -> 99.56 | Helped |
| Increase sampled candidate models beyond B=4 (ablation) | Search-Softmax · LFW, MS-Celeb-1M-v1c-R | accuracy: 99.78 -> 99.79 (B=8); 99.78 -> 99.78 (B=16) | No clear effect |
| Use reward-guided search rather than randomly choosing the modulating factor | Search-Softmax · Trillion-Pairs identification, CASIA-WebFace-R | TPR@FAR=1e-3: 27.70 -> 29.23 | Helped |
| Use reward-guided search rather than randomly choosing the modulating factor | Search-Softmax · Trillion-Pairs verification, CASIA-WebFace-R | TPR@FAR=1e-9: 27.28 -> 28.49 | Helped |
Closest papers by meaning
- Loss Function Discovery for Object Detection via Convergence-Simulation Driven Search (ICLR 2021 · similarity 0.90)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.90)
- CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018 · similarity 0.87)
- Learning Towards The Largest Margins (ICLR 2022 · similarity 0.87)
- GIF: Generative Inspiration for Face Recognition at Scale (CVPR 2025 · similarity 0.84)
- Learning with Fitzpatrick Losses (NeurIPS 2024 · 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.