Research map / Learning theory and algorithms / Prediction and decision losses
AM-LFS: AutoML for Loss Function Search
Chuming Li, Xin Yuan, Chen Lin, Minghao Guo, Wei Wu, Junjie Yan, Wanli Ouyang
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 |
|---|---|---|---|
| replace cross-entropy with AM-LFS | ResNet-20 · CIFAR-10 | top-1 error (%): 8.75 -> 6.92 | Improved |
| replace L2T-DLF with AM-LFS | ResNet-20 · CIFAR-10 | top-1 error (%): 7.63 -> 6.92 | Improved |
| replace cross-entropy with AM-LFS under 10% label noise | ResNet-20 · CIFAR-10 (10% noise) | top-1 error (%): 12.05 -> 10.05 | Improved |
| replace cross-entropy with AM-LFS under 20% label noise | ResNet-20 · CIFAR-10 (20% noise) | top-1 error (%): 15.05 -> 12.73 | Improved |
| replace ArcFace loss with AM-LFS | ResNet-20 · MegaFace | rank1@1e6 (%): 72.4 -> 73.5 | Improved |
| add AM-LFS to MGN(RK) | MGN(RK) · Market-1501 | mAP (%): 94.2 -> 94.6 | Improved |
| add AM-LFS to MGN(RK) | MGN(RK) · Market-1501 | rank-1 (%): 96.6 -> 96.1 | Hurt |
| add AM-LFS to MGN(RK) | MGN(RK) · DukeMTMC-reID | mAP (%): 88.6 -> 90.1 | Improved |
| add the search space with random sampling (no distribution update) to SphereReID (ablation) | SphereReID · Market-1501 | mAP (%): 83.6 -> 84.0 | Improved |
| add optimization of the loss distribution on top of fixed-distribution sampling (ablation) | AM-LFS (SphereReID) · Market-1501 | mAP (%): 84.0 -> 84.4 | Improved |
| increase number of samples B from 4 to 32 (ablation) | AM-LFS (SphereReID) · Market-1501 | mAP (%): 83.6 -> 84.4 | Improved |
| increase number of samples B from 4 to 32 (ablation) | AM-LFS (SphereReID) · DukeMTMC-reID | mAP (%): 68.4 -> 69.8 | Improved |
Closest papers by meaning
- AutoLoss-Zero: Searching Loss Functions From Scratch for Generic Tasks (CVPR 2022 · similarity 0.89)
- Loss Function Discovery for Object Detection via Convergence-Simulation Driven Search (ICLR 2021 · similarity 0.89)
- Loss Function Search for Face Recognition (ICML 2020 · similarity 0.88)
- AutoMMLab: Automatically Generating Deployable Models from Language Instructions for Computer Vision Tasks (AAAI 2025 · similarity 0.82)
- Auto-Prox: Training-Free Vision Transformer Architecture Search via Automatic Proxy Discovery (AAAI 2024 · similarity 0.82)
- Leaving the Nest: Going beyond Local Loss Functions for Predict-Then-Optimize (AAAI 2024 · similarity 0.81)
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.
