Research map / Large language models / Language, safety and interpretability
Improving Commonsense Causal Reasoning by Adversarial Training and Data Augmentation
Ieva Staliūnaitė, Philip John Gorinski, Ignacio Iacobacci
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 adversarially perturbed training items | RoBERTa-large COPA classifier · COPA Dev | accuracy: 86.21 -> 87.35 | Improved |
| Add adversarially perturbed training items | RoBERTa-large COPA classifier · COPA Test | accuracy: 85.69 -> 88.19 | Improved |
| Add adversarially perturbed training items | RoBERTa-large COPA classifier · Balanced COPA Dev | accuracy: 81.72 -> 82.84 | Improved |
| Add 400 random web-text distractor examples | RoBERTa-large COPA classifier · COPA Dev | accuracy: 86.21 -> 86.73 | Improved |
| Add overlap-based web-text distractor examples | RoBERTa-large COPA classifier · COPA Test | accuracy: 85.69 -> 88.13 | Improved |
| Add GPT-2-generated distractor examples | RoBERTa-large COPA classifier · COPA Test | accuracy: 85.69 -> 90.24 | Improved |
| Add GPT-2-generated distractor examples | RoBERTa-large COPA classifier · Balanced COPA Dev | accuracy: 81.72 -> 83.94 | Improved |
| Add random, overlap, and GPT-2-generated examples together | RoBERTa-large COPA classifier · COPA Dev | accuracy: 86.21 -> 87.75 | Improved |
| Add random, overlap, and GPT-2-generated examples together | RoBERTa-large COPA classifier · COPA Test | accuracy: 85.69 -> 89.44 | Improved |
| Add random, overlap, and GPT-2-generated examples together | RoBERTa-large COPA classifier · Balanced COPA Dev | accuracy: 81.72 -> 83.78 | Improved |
| Train on generated examples without adding the original COPA training data | RoBERTa-large COPA classifier · COPA Dev | accuracy: 86.21 -> 68.51; 66.01; 71.25; 76.85 | Hurt |
| Train on generated examples without adding the original COPA training data | RoBERTa-large COPA classifier · COPA Test | accuracy: 85.69 -> 63.63; 65.19; 71.31; 77.13 | Hurt |
Closest papers by meaning
- Can Large Language Models Infer Causation from Correlation? (ICLR 2024 · similarity 0.88)
- Rethinking the Development of Large Language Models from the Causal Perspective: A Legal Text Prediction Case Study (AAAI 2024 · similarity 0.87)
- Unveiling Causal Reasoning in Large Language Models: Reality or Mirage? (NeurIPS 2024 · similarity 0.87)
- COLD: Causal reasOning in cLosed Daily activities (NeurIPS 2024 · similarity 0.86)
- Eliciting Causal Abilities in Large Language Models for Reasoning Tasks (AAAI 2025 · similarity 0.86)
- Causality can systematically address the monsters under the bench(marks) (NeurIPS 2026 · similarity 0.85)
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.
