Research map / Architectures and efficiency / Neural architecture search
AutoLoss-Zero: Searching Loss Functions From Scratch for Generic Tasks
Check 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 |
|---|---|---|---|
| search loss functions from scratch with primitive operators for mIoU training | DeepLabv3+ ResNet-101 · PASCAL VOC 2012 | mIoU: 80.7 vs 81.0 | Mixed |
| search loss functions from scratch with primitive operators for mIoU training | DeepLabv3+ ResNet-101 · PASCAL VOC 2012 | BIoU: 74.1 vs 73.4 | Helped |
| search loss functions from scratch with primitive operators for mIoU training | DeepLabv3+ ResNet-101 · PASCAL VOC 2012 | BF1: 66.0 vs 68.9 | Hurt |
| search all four loss branches from scratch | Faster R-CNN ResNet-50 FPN · COCO | mAP: 38.1 vs 38.5 | Mixed |
| search loss functions from scratch | Mask R-CNN ResNet-50 FPN · COCO | mAP: 34.8 vs 34.7 | Helped |
| search the pose loss from scratch | Pose estimation network ResNet-50 · COCO | keypoint mAP: 72.0 vs 71.5 | Helped |
| add the loss-rejection protocol to naïve evolution (ablation) | AutoLoss-Zero search · Object detection search (48 hours) | Speed-Up: 1× -> ∼700× | Helped |
| add gradient-equivalence checking after loss rejection (ablation) | AutoLoss-Zero search · Object detection search (48 hours) | Speed-Up: ∼700× -> ∼1000× | Helped |
| stop training when loss values are invalid (ablation) | AutoLoss-Zero search · Object detection search (48 hours) | Speed-Up: ∼1000× -> ∼5000× | Helped |
| use naïve evolution without loss-rejection protocol (ablation) | object detection loss search with four branches · COCO | mAP | Hurt |
Closest papers by meaning
- Loss Function Discovery for Object Detection via Convergence-Simulation Driven Search (ICLR 2021 · similarity 0.89)
- Auto Seg-Loss: Searching Metric Surrogates for Semantic Segmentation (ICLR 2021 · similarity 0.86)
- AutoLoss: Learning Discrete Schedule for Alternate Optimization (ICLR 2019 · similarity 0.85)
- Automated Loss function Search for Class-imbalanced Node Classification (ICML 2024 · similarity 0.83)
- Loss Landscape Characterization of Neural Networks without Over-Parametrization (NeurIPS 2024 · similarity 0.83)
- Auto-Prox: Training-Free Vision Transformer Architecture Search via Automatic Proxy Discovery (AAAI 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.