AI Paper Atlas

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

Breaking the Adversarial Robustness-Performance Trade-off in Text Classification via Manifold Purification

Chenhao Dang, Jing Ma

AAAI 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Robust overfitting and regularization

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 negative log-likelihood loss (density estimation) to MC2F training (ablation)MC2F · AGNews TextFooleraccuracy under attack (Aua%): 32.6 -> 53.8Improved
add topological regularization loss to MC2F training (ablation)MC2F · AGNews TextFooleraccuracy under attack (Aua%): 32.9 -> 53.8Improved
add causal and semantic (Fisher-Rao) loss to MC2F training (ablation)MC2F · AGNews TextFooleraccuracy under attack (Aua%): 48.6 -> 53.8Improved
add topological regularization loss to MC2F training (ablation)MC2F · AGNews clean accuracyClean%: 93.41 -> 95.13Improved
add causal loss to MC2F training (ablation)MC2F · AGNews clean accuracyClean%: 94.76 -> 95.13No clear effect
replace fine-tuned BERT with MC2F purification defenseBERT-base · AGNews TextFooleraccuracy under attack (Aua%): 14.7 -> 53.8Improved
use MC2F instead of Subspace Defense (SD)BERT-base · AGNews BERT-Attackaccuracy under attack (Aua%): 38.6 -> 45.3Improved
use MC2F instead of Subspace Defense (SD)BERT-base · YELP clean test setClean%: 93.45 -> 95.26Improved
use MC2F instead of Subspace Defense (SD)BERT-base · YELP BERT-Attackaverage queries per successful attack: 320.7 -> 586.4Improved
use MC2F instead of standard fine-tuningBERT-base · SST-2 clean test setClean%: 92.71 -> 92.71No clear effect

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