Research map / Label-efficient and robust learning / Recognition and long-tail
Data Augmentations for Improved (Large) Language Model Generalization
Amir Feder, Yoav Wald, Claudia Shi, Suchi Saria, David Blei
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 counterfactual augmentation (CATO B) instead of observational training | logistic regression · CeBAB-Spurious | accuracy: 0.64 -> 0.75 | Improved |
| add naive LLM augmentation | logistic regression · CeBAB-Spurious | accuracy: 0.64 -> 0.62 | Hurt |
| add naive LLM augmentation | logistic regression · CeBAB | accuracy: 0.85 -> 0.80 | Hurt |
| use conditional augmentation without diff-in-diff | logistic regression · CeBAB-Spurious | accuracy: 0.64 -> 0.70 | Improved |
| add diff-in-diff counterfactual augmentation (Causal Aug.) | PubMED BERT · i2b2-2006 OOD demographic traits | F1: 66.92 -> 71.88 | Improved |
| add causal augmentation | PubMED BERT · i2b2-2006 OOD demographic traits | precision: 53.32 -> 59.76 | Improved |
| add naive augmentation | PubMED BERT · i2b2-2006 OOD demographic traits | F1: 66.92 -> 66.52 | No clear effect |
| add naive augmentation | PubMED BERT · MIMIC-III ID demographic traits | F1: 79.34 -> 80.39 | Improved |
| add causal augmentation | PubMED BERT · MIMIC-III ID demographic traits | F1: 79.34 -> 79.73 | No clear effect |
| add re-weighting | PubMED BERT · i2b2-2006 OOD demographic traits | F1: 66.92 -> 70.02 | Improved |
| replace PubMED BERT with GPT3 as baseline model | observational baseline · i2b2-2006 OOD demographic traits | F1: 66.92 -> 63.98 | Hurt |
| add causal augmentation | PubMED BERT · i2b2-2010 OOD clinical condition | precision: more than 7% absolute | Improved |
Closest papers by meaning
- Robustness to Spurious Correlations in Text Classification via Automatically Generated Counterfactuals (AAAI 2021 · similarity 0.90)
- Learning The Difference That Makes A Difference With Counterfactually-Augmented Data (ICLR 2020 · similarity 0.90)
- Iterative Counterfactual Data Augmentation (AAAI 2025 · similarity 0.84)
- A General Search-Based Framework for Generating Textual Counterfactual Explanations (AAAI 2024 · similarity 0.84)
- Not All LLM-Generated Data Are Equal: Rethinking Data Weighting in Text Classification (ICLR 2025 · similarity 0.84)
- Compositional Generalization for Multi-Label Text Classification: A Data-Augmentation Approach (AAAI 2024 · 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.
