Research map / Architectures and efficiency / Neural architecture search
Online Hyper-Parameter Learning for Auto-Augmentation Strategy
Chen Lin, Minghao Guo, Chuming Li, Xin Yuan, Wei Wu, Junjie Yan, Dahua Lin, 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 |
|---|---|---|---|
| use OHL-Auto-Aug instead of baseline augmentation | ResNet-18 · CIFAR-10 test | error rate (%): 4.66 -> 3.29 | Improved |
| use OHL-Auto-Aug instead of Auto-Augment | WideResNet-28-10 · CIFAR-10 test | error rate (%): 2.68 -> 2.61 | Improved |
| use OHL-Auto-Aug instead of Auto-Augment | AmoebaNet-B(6,128) · CIFAR-10 test | error rate (%): 1.75 -> 1.89 | Hurt |
| use OHL-Auto-Aug instead of our-implementation baseline | AmoebaNet-B(6,128) · CIFAR-10 test | error rate (%): 3.4 -> 1.89 | Improved |
| use OHL-Auto-Aug instead of Auto-Augment | ResNet-50 · ImageNet val | top-1 error (%): 22.37 -> 21.07 | Improved |
| use OHL-Auto-Aug instead of baseline | SE-ResNeXt-101 · ImageNet val | top-1 error (%): 20.70 -> 19.30 | Improved |
| use OHL-Auto-Aug instead of Auto-Augment | SE-ResNeXt-101 · ImageNet val | top-5 error (%): 5.23 -> 4.57 | Hurt |
| replace Auto-Augment search with online learning | search cost · CIFAR-10 search cost | search iterations: 7.03x10^6 -> 1.17x10^5 | Improved |
| replace Auto-Augment search with online learning | search cost · ImageNet search cost | search iterations: 1.76x10^7 -> 7.5x10^5 | Improved |
| learn augmentation distribution (eta>0) instead of uniform sampling (eta=0) (ablation) | OHL-Auto-Aug ResNet-18 · CIFAR-10 test | accuracy | Improved |
| increase trajectory samples from 1 to 8 (ablation) | OHL-Auto-Aug ResNet-18 · CIFAR-10 test | accuracy | Improved |
| increase trajectory samples beyond 8 (ablation) | OHL-Auto-Aug ResNet-18 · CIFAR-10 test | accuracy | No clear effect |
Closest papers by meaning
- OnlineAugment: Online Data Augmentation with Less Domain Knowledge (ECCV 2020 · similarity 0.88)
- Faster AutoAugment: Learning Augmentation Strategies Using Backpropagation (ECCV 2020 · similarity 0.88)
- Improving Auto-Augment via Augmentation-Wise Weight Sharing (NeurIPS 2020 · similarity 0.87)
- Adversarial AutoAugment (ICLR 2020 · similarity 0.86)
- AutoAugment: Learning Augmentation Strategies From Data (CVPR 2019 · similarity 0.85)
- LLM-AutoDA: Large Language Model-Driven Automatic Data Augmentation for Long-tailed Problems (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.
