Research map / Label-efficient and robust learning / Recognition and long-tail
Post-training Iterative Hierarchical Data Augmentation for Deep Networks
Adil Khan, Khadija Fraz
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 |
|---|---|---|---|
| add IHDA fine-tuning on feature-space augmentations | ResNet-110 · CIFAR-10 test | error (%): 6.45 -> 2.78 | Improved |
| add IHDA fine-tuning | ResNet-110 · CIFAR-100 test | error (%): 28.57 -> 18.45 | Improved |
| add standard input-space augmentation (IHDA+ vs IHDA) (ablation) | IHDA ResNet-110 · CIFAR-10 test | error (%): 2.78 -> 2.72 | Improved |
| IHDA instead of AAA on Wide-ResNet-28-10 | Wide-ResNet-28-10 · CIFAR-10 test | error (%): 1.90 -> 2.19 | Hurt |
| fine-tune all subsequent layers instead of only the predictor (ablation) | IHDA ResNet-110 · CIFAR-10 test | error (%): 6.01 -> 2.78 | Improved |
| augment at every level (p=0) versus level dropout p=0.55 (ablation) | IHDA ResNet-110 · CIFAR-10 test | error (%): 6.87 -> 2.78 | Improved |
| decrease dropout probability for deeper levels instead of same p (ablation) | IHDA ResNet-110 · CIFAR-10 test | error (%): 5.01 -> 2.78 | Improved |
| select source points by potential in hard regions instead of random (ablation) | IHDA ResNet-110 · CIFAR-10 test | error (%): 5.73 -> 2.78 | Improved |
| select source points by potential instead of random (ablation) | IHDA ResNet-110 · CIFAR-100 test | error (%): 26.91 -> 18.45 | Improved |
| IHDA versus training baseline for same extra epochs (ablation) | ResNet-110 · CIFAR-10 test | error (%): 6.33 -> 2.78 | Improved |
| add IHDA to MLP classifier | SHL network · SHL test | accuracy (%): 83% -> 92% ± 0.05% | Improved |
| combine IHDA+ with AutoAugment policies | Wide-ResNet-28-10 · CIFAR-10 test | error (%): 2.11 -> 1.92 | Improved |
Closest papers by meaning
- Data-Efficient Augmentation for Training Neural Networks (NeurIPS 2022 · similarity 0.85)
- Scalable Hierarchical Deep Neural Network for Time Series Analysis in Wearable Sensor-based Human Activity Recognition (AAAI 2025 · similarity 0.84)
- Hierarchical Features Matter: A Deep Exploration of Progressive Parameterization Method for Dataset Distillation (CVPR 2025 · similarity 0.82)
- Improved Generalization of Weight Space Networks via Augmentations (ICML 2024 · similarity 0.82)
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · similarity 0.82)
- How Deep Networks Learn Sparse and Hierarchical Data: the Sparse Random Hierarchy Model (ICML 2024 · 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.
