Research map / Label-efficient and robust learning / Recognition and long-tail
Reweighting Augmented Samples by Minimizing the Maximal Expected Loss
Mingyang Yi, Lu Hou, Lifeng Shang, Xin Jiang, Qun Liu, Zhi-Ming Ma
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 loss-based MMEL weights instead of uniform weights | ResNet56 · CIFAR-10 | accuracy: 94.62($\pm$0.18) -> 94.85($\pm$0.15) | Improved |
| Use loss-based MMEL weights instead of uniform weights | ResNet56 · CIFAR-100 | accuracy: 73.44($\pm$0.13) -> 74.20($\pm$0.24) | Improved |
| Apply MMEL-H with standard image augmentation | ResNet · ImageNet | accuracy: +0.72; +1.08; +0.38 | Improved |
| Use soft consistency loss rather than the hard-label loss | ResNet · ImageNet | accuracy: -1.24; -0.97; -1.33 | Hurt |
| Add MMEL reweighting with augmented text samples | BERT_BASE · GLUE dev Avg | Avg: 83.5($\pm$0.27) -> 84.5($\pm$0.11) | Improved |
| Add MMEL reweighting with augmented text samples | BERT_BASE · GLUE test Avg | Avg: 79.6($\pm$0.09) -> 80.5($\pm$0.17) | Improved |
| Use teacher-predicted labels for augmented samples instead of original labels (ablation) | MMEL-H · GLUE dev | Avg: 78.1 -> 84.5 | Improved |
| Use teacher-predicted labels for augmented samples instead of original labels (ablation) | MMEL-S · GLUE dev | Avg: 80.0 -> 84.7 | Improved |
| Increase augmented samples from 2 to 5 (ablation) | MMEL-H · CIFAR-10, ResNet56 | accuracy: 94.29($\pm$0.08) -> 94.78($\pm$0.09) | Improved |
| Increase augmented samples from 10 to 20 (ablation) | MMEL-H · CIFAR-100, ResNet56 | accuracy: 74.20($\pm$0.24) -> 74.10($\pm$0.05) | No clear effect |
| Use beam search rather than greedy search to generate augmented text (ablation) | MMEL · GLUE dev | Avg | No clear effect |
| Generate the two sentences separately rather than together (ablation) | MMEL · GLUE dev | Avg | Mixed |
Closest papers by meaning
- Thumb on the Scale: Optimal Loss Weighting in Last Layer Retraining (NeurIPS 2025 · similarity 0.87)
- AugMax: Adversarial Composition of Random Augmentations for Robust Training (NeurIPS 2021 · similarity 0.86)
- Tradeoffs in Data Augmentation: An Empirical Study (ICLR 2021 · similarity 0.86)
- Dynamic Loss-Based Sample Reweighting for Improved Large Language Model Pretraining (ICLR 2025 · similarity 0.86)
- A Bayesian Data Augmentation Approach for Learning Deep Models (NeurIPS 2017 · similarity 0.86)
- Augment Your Batch: Improving Generalization Through Instance Repetition (CVPR 2020 · 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.
