AI Paper Atlas

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

AAAI 2021 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations · 16 citations (OpenAlex)

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 changedWhereMeasuredEffect
Add adversarially perturbed training itemsRoBERTa-large COPA classifier · COPA Devaccuracy: 86.21 -> 87.35Improved
Add adversarially perturbed training itemsRoBERTa-large COPA classifier · COPA Testaccuracy: 85.69 -> 88.19Improved
Add adversarially perturbed training itemsRoBERTa-large COPA classifier · Balanced COPA Devaccuracy: 81.72 -> 82.84Improved
Add 400 random web-text distractor examplesRoBERTa-large COPA classifier · COPA Devaccuracy: 86.21 -> 86.73Improved
Add overlap-based web-text distractor examplesRoBERTa-large COPA classifier · COPA Testaccuracy: 85.69 -> 88.13Improved
Add GPT-2-generated distractor examplesRoBERTa-large COPA classifier · COPA Testaccuracy: 85.69 -> 90.24Improved
Add GPT-2-generated distractor examplesRoBERTa-large COPA classifier · Balanced COPA Devaccuracy: 81.72 -> 83.94Improved
Add random, overlap, and GPT-2-generated examples togetherRoBERTa-large COPA classifier · COPA Devaccuracy: 86.21 -> 87.75Improved
Add random, overlap, and GPT-2-generated examples togetherRoBERTa-large COPA classifier · COPA Testaccuracy: 85.69 -> 89.44Improved
Add random, overlap, and GPT-2-generated examples togetherRoBERTa-large COPA classifier · Balanced COPA Devaccuracy: 81.72 -> 83.78Improved
Train on generated examples without adding the original COPA training dataRoBERTa-large COPA classifier · COPA Devaccuracy: 86.21 -> 68.51; 66.01; 71.25; 76.85Hurt
Train on generated examples without adding the original COPA training dataRoBERTa-large COPA classifier · COPA Testaccuracy: 85.69 -> 63.63; 65.19; 71.31; 77.13Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy