Research map / Label-efficient and robust learning / Recognition and long-tail
CoDA: Contrast-enhanced and Diversity-promoting Data Augmentation for Natural Language Understanding
Yanru Qu, Dinghan Shen, Yelong Shen, Sandra Sajeev, Weizhu Chen, Jiawei Han
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 |
|---|---|---|---|
| Stack adversarial training over back-translation rather than use either transformation alone (ablation) | RoBERTa-base · MNLI-m dev | accuracy: 88.5 -> 88.8 | Improved |
| Mix back-translation and adversarial samples instead of using back-translation alone (ablation) | RoBERTa-base · MNLI-m dev | accuracy: 88.5 -> 88.6 | Improved |
| Add cutoff to the back-translation and adversarial stack (ablation) | RoBERTa-base · MNLI-m dev | accuracy: 88.8 -> 88.5 | Hurt |
| Put cutoff after the back-translation and adversarial stack (ablation) | RoBERTa-base · MNLI-m dev | accuracy: 88.8 -> 88.4 | Hurt |
| Randomly select among back-translation, cutoff, and adversarial training (ablation) | RoBERTa-base · MNLI-m dev | accuracy: 88.8 -> 88.4 | Hurt |
| Use CoDA instead of standard fine-tuning | RoBERTa-large · GLUE dev | average: 88.9 -> 91.1 | Improved |
| Use the proposed contrastive objective rather than MoCo | RoBERTa-base · QNLI dev | accuracy: 93.3 -> 93.6 | Improved |
| Use the proposed contrastive objective rather than SupCon | RoBERTa-base · RTE dev | accuracy: 80.5 -> 82.0 | Improved |
| Increase memory-bank size (ablation) | CoDA contrastive objective · MNLI-m dev | accuracy | Improved |
| Apply contrastive regularization to different data augmentation approaches | RoBERTa-base · QNLI | performance | Improved |
| Use CoDA rather than back-translation or adversarial training alone | RoBERTa-base · MNLI and QNLI low-resource settings | performance | Improved |
Later papers that compared with this method
4 later accepted papers report CoDA as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Heuristic Self-Paced Learning for Domain Adaptive Semantic Segmentation under Adverse Conditions (CVPR 2026 · better on 2 of 2 shared results)
- Exploring Weather-aware Aggregation and Adaptation for Semantic Segmentation under Adverse Conditions (ICCV 2025 · better on 3 of 3 shared results)
- Semantic Feature Learning for Universal Unsupervised Cross-Domain Retrieval (NeurIPS 2024 · better on 2 of 2 shared results)
- DiDA: Disambiguated Domain Alignment for Cross-Domain Retrieval with Partial Labels (AAAI 2024 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Contrastive Visual Data Augmentation (ICML 2025 · similarity 0.85)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 2025 · similarity 0.84)
- Compositional Generalization for Multi-Label Text Classification: A Data-Augmentation Approach (AAAI 2024 · similarity 0.84)
- Progressive Compositionality in Text-to-Image Generative Models (ICLR 2025 · similarity 0.84)
- GRACE: Generative Representation Learning via Contrastive Policy Optimization (ICLR 2026 · similarity 0.84)
- Synthetic continued pretraining (ICLR 2025 · similarity 0.84)
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.
