Research map / Reinforcement learning / Visual and representation RL
Deep 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 a multi-layer augmentation policy from scratch without default augmentations | DeepAA · ImageNet | Top-1 test accuracy: 78.07 -> 78.30 | Helped |
| Search a multi-layer augmentation policy from scratch without default augmentations | DeepAA · ImageNet | Top-1 test accuracy: 80.6 -> 81.32 | Helped |
| Search a multi-layer policy rather than use TA(Wide) | WRN-28-10 · CIFAR-10 | Top-1 test accuracy: 97.46 -> 97.56 | Helped |
| Search a multi-layer policy rather than use TA(Wide) | Shake-Shake-2x96d · CIFAR-10 | Top-1 test accuracy: 98.21 -> 98.11 | Hurt |
| Search a multi-layer policy rather than use TA(Wide) | WRN-28-10 · CIFAR-100 | Top-1 test accuracy: 84.33 -> 84.02 | Hurt |
| Search a multi-layer policy rather than use TA(Wide) | Shake-Shake-2x96d · CIFAR-100 | Top-1 test accuracy: 86.19 -> 85.19 | Hurt |
| Add Batch Augmentation using the searched DeepAA policy | DeepAA with WRN-28-10 · CIFAR-10 | Top-1 test accuracy: 98.06 ± 0.23 -> 98.21 ± 0.14 | Helped |
| Add Batch Augmentation using the searched DeepAA policy | DeepAA with WRN-28-10 · CIFAR-100 | Top-1 test accuracy: 85.40 ± 0.15 -> 85.61 ± 0.17 | Helped |
| Add augmentation layers from 1 to 5 (ablation) | DeepAA · CIFAR-10 | Top-1 test accuracy: 96.3 ± 0.21 -> 97.56 ± 0.14 | Helped |
| Add a sixth augmentation layer after five layers (ablation) | DeepAA · CIFAR-10 | Top-1 test accuracy: 97.56 ± 0.14 -> 97.6 ± 0.12 | No clear effect |
| Add augmentation layers from 1 to 5 (ablation) | DeepAA · CIFAR-100 | Top-1 test accuracy: 80.9 ± 0.31 -> 84.02 ± 0.18 | Helped |
| Add a sixth augmentation layer after five layers (ablation) | DeepAA · CIFAR-100 | Top-1 test accuracy: 84.02 ± 0.18 -> 84.0 ± 0.19 | No clear effect |
Closest papers by meaning
- AutoAugment: Learning Augmentation Strategies From Data (CVPR 2019 · similarity 0.88)
- Fast AutoAugment (NeurIPS 2019 · similarity 0.88)
- AdaAug: Learning Class- and Instance-adaptive Data Augmentation Policies (ICLR 2022 · similarity 0.87)
- RandAugment: Practical Automated Data Augmentation with a Reduced Search Space (NeurIPS 2020 · similarity 0.86)
- Adversarial AutoAugment (ICLR 2020 · similarity 0.86)
- Revisiting Data Augmentation in Deep Reinforcement Learning (ICLR 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.