Research map / Architectures and efficiency / Neural architecture search
Fast AutoAugment
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 |
|---|---|---|---|
| Search policies by density matching and Bayesian optimization without retraining a model for each candidate | Fast AutoAugment · CIFAR-10 policy search | GPU hours: 5000 -> 3.5 | Helped |
| Search policies by density matching and Bayesian optimization without retraining a model for each candidate | Fast AutoAugment · SVHN policy search | GPU hours: 1000 -> 1.5 | Helped |
| Search policies by density matching and Bayesian optimization without retraining a model for each candidate | Fast AutoAugment · ImageNet policy search | GPU hours: 15000 -> 450 | Helped |
| Use Fast AutoAugment instead of AutoAugment | ResNet-200 · ImageNet validation | Top-1 error rate (%): 20.00 -> 19.4 | Helped |
| Use Fast AutoAugment instead of AutoAugment | ResNet-200 · ImageNet validation | Top-5 error rate (%): 5.0 -> 4.7 | Helped |
| Use Fast AutoAugment instead of AutoAugment | ResNet-50 · ImageNet validation | Top-1 error rate (%): 22.4 -> 22.4 | No clear effect |
| Use Fast AutoAugment instead of AutoAugment | ResNet-50 · ImageNet validation | Top-5 error rate (%): 6.2 -> 6.3 | Hurt |
| Use Fast AutoAugment policies rather than baseline preprocessing | Wide-ResNet-40-2 · CIFAR-10 test | error rate (%): 5.3 -> 3.6 | Helped |
| Use Fast AutoAugment policies rather than baseline preprocessing | Wide-ResNet-40-2 · CIFAR-100 test | error rate (%): 26.0 -> 20.6 | Helped |
| Use Fast AutoAugment policies rather than baseline preprocessing | Wide-ResNet-28-10 · SVHN test | error rate (%): 1.5 -> 1.1 | Helped |
| Increase the number of searched sub-policies used for training (ablation) | Fast AutoAugment · CIFAR-10 and CIFAR-100 validation | error rate | Helped |
| Transfer policies searched using a smaller model rather than search directly for a larger model (ablation) | Fast AutoAugment · CIFAR-10 and CIFAR-100 | error rate | Mixed |
Closest papers by meaning
- AutoAugment: Learning Augmentation Strategies From Data (CVPR 2019 · similarity 0.89)
- Deep AutoAugment (ICLR 2022 · similarity 0.88)
- Adversarial AutoAugment (ICLR 2020 · similarity 0.87)
- RandAugment: Practical Automated Data Augmentation with a Reduced Search Space (NeurIPS 2020 · similarity 0.87)
- Your Image is My Video: Reshaping the Receptive Field via Image-To-Video Differentiable AutoAugmentation and Fusion (CVPR 2024 · similarity 0.81)
- AugGen: Synthetic Augmentation using Diffusion Models Can Improve Recognition (NeurIPS 2025 · 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.