AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Unifying Model Explainability and Robustness for Joint Text Classification and Rationale Extraction

Dongfang Li, Baotian Hu, Qingcai Chen, Tujie Xu, Jingcong Tao, Yunan Zhang

AAAI 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training with feature consistency · 10 citations (OpenAlex)

Read the paper

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
Use AT-BMC with BERT-base instead of the strongest listed baselinetext classification and rationale extraction · Movie ReviewsToken F1: 46.3 -> 50.6Improved
Use AT-BMC with BERT-base instead of the strongest listed baselinetext classification and rationale extraction · MultiRCToken F1: 54.0 -> 57.3Improved
Use AT-BMC with BERT-base instead of vanilla BERTMovie Reviews · TextFooler attackattack success rate: 98.89 -> 57.69Improved
Use AT-BMC with BERT-base instead of vanilla BERTMovie Reviews · TextBugger attackattack success rate: 90.00 -> 26.92Improved
Use AT-BMC with BERT-base instead of vanilla BERTMovie Reviews · PWWS attackattack success rate: 96.67 -> 26.92Improved
Add label embedding to condition rationale extraction on the predicted label (ablation)AT-BMC · Movie Reviews and MultiRCToken F1: 49.8±0.51 -> 50.6±0.54; 56.2±0.47 -> 57.3±0.34Improved
Add discrete adversarial examples outside rationale spans (ablation)AT-BMC · Movie ReviewsToken F1: 48.0±0.34 -> 50.6±0.54Improved
Add span match loss (ablation)AT-BMC · Movie ReviewsToken F1: 47.4±0.45 -> 50.6±0.54Improved
Add virtual adversarial training in embedding space (ablation)AT-BMC · MultiRCaccuracy: 63.6±0.47 -> 67.7±0.49Improved
Use mixed adversarial training rather than removing it (ablation)AT-BMC · MultiRCaccuracy: 63.9±0.41 -> 67.7±0.49Improved

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