Research map / Architectures and efficiency / Neural architecture search
DADA: Differentiable Automatic Data Augmentation
Yonggang Li, Guosheng Hu, Yongtao Wang, Timothy Hospedales, Neil M. Robertson, Yongxin Yang
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 |
|---|---|---|---|
| search policy on full CIFAR-10 instead of reduced set (ablation) | DADA · CIFAR-10 WRN-40-2 | test error: 3.61 -> 3.53 | Improved |
| search policy on full CIFAR-10 instead of reduced set (ablation) | DADA · CIFAR-10 WRN-28-10 | test error: 2.73 -> 2.64 | Improved |
| use unbiased RELAX estimator instead of Gumbel-Softmax estimator (ablation) | DADA · CIFAR-10 | test error | Improved |
| apply DADA policy versus no augmentation policy | WRN-40-2 · CIFAR-10 | test error: 5.3 -> 3.6 | Improved |
| apply DADA policy versus no augmentation policy | ResNet-50 · ImageNet | top-1 error: 23.7 -> 22.5 | Improved |
| pretrain backbone with DADA policy | RetinaNet ResNet-50 · COCO | box AP: 35.9 -> 36.6 | Improved |
| pretrain backbone with DADA policy | Faster R-CNN ResNet-50 · COCO | box AP: 36.6 -> 37.2 | Improved |
| pretrain backbone with DADA policy | Mask R-CNN ResNet-50 · COCO | mask AP: 34.2 -> 34.5 | Improved |
| DADA instead of Fast AA for search | policy search · CIFAR-10 | GPU hours: 3.5 -> 0.1 | Improved |
| DADA versus AA on CIFAR-100 WRN-28-10 | policy search · CIFAR-100 | test error: 17.1 -> 17.5 | Hurt |
| DADA versus Fast AA on full-dataset search | policy search · CIFAR-10 | test error | Mixed |
Closest papers by meaning
- Faster AutoAugment: Learning Augmentation Strategies Using Backpropagation (ECCV 2020 · similarity 0.88)
- Population Based Augmentation: Efficient Learning of Augmentation Policy Schedules (ICML 2019 · similarity 0.87)
- Direct Differentiable Augmentation Search (ICCV 2021 · similarity 0.86)
- Revisiting Data Augmentation in Deep Reinforcement Learning (ICLR 2024 · similarity 0.86)
- AdaAug: Learning Class- and Instance-adaptive Data Augmentation Policies (ICLR 2022 · similarity 0.86)
- Learning Better with Less: Effective Augmentation for Sample-Efficient Visual Reinforcement Learning (NeurIPS 2023 · similarity 0.86)
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.
