Adversarial Reweighting

Adversarial reweighting is a technique used to improve the performance and fairness of machine learning models by strategically adjusting the influence of individual data points during training. Current research focuses on applying this method to address various challenges, including improving robustness against adversarial attacks, mitigating bias stemming from imbalanced datasets, and enhancing domain adaptation performance. This approach leverages adversarial training frameworks, often incorporating techniques like Wasserstein distance or class activation mapping to guide the reweighting process, ultimately aiming for more accurate, robust, and equitable models. The impact of this research extends to various applications, from improving the reliability of image classification systems to promoting fairness in speaker verification and other decision-making systems.

Papers

June 2, 2024

Improving Accuracy-robustness Trade-off via Pixel Reweighted Adversarial Training
Jiacheng Zhang, Feng Liu, Dawei Zhou, Jingfeng Zhang, Tongliang Liu
Adversarial Example Adversarial Training Robust Classification Accuracy Robustness Adversarial Reweighting

April 26, 2024

Adversarial Reweighting with $\alpha$-Power Maximization for Domain Adaptation
Xiang Gu, Xi Yu, Yan Yang, Jian Sun, Zongben Xu
Domain Adaptation Transferable Time Series Optimal Power Adversarial Reweighting

November 21, 2023

Adversarial Reweighting Guided by Wasserstein Distance for Bias Mitigation
Xuan Zhao, Simone Fabbrizzi, Paula Reyero Lobo, Siamak Ghodsi, Klaus Broelemann, Steffen Staab, Gjergji Kasneci
Bias Mitigation Wasserstein Distance Fairness Aware Machine Representation Bias Adversarial Reweighting

August 1, 2023

Doubly Robust Instance-Reweighted Adversarial Training
Daouda Sow, Sen Lin, Zhangyang Wang, Yingbin Liang
Adversarial Attack Adversarial Training Robust Version Adversarial Data Importance Weight Adversarial Reweighting

July 15, 2022

Adversarial Reweighting for Speaker Verification Fairness
Minho Jin, Chelsea J. -T. Ju, Zeya Chen, Yi-Chieh Liu, Jasha Droppo, Andreas Stolcke
Adversarial Learning Speaker Verification Pairwise Learning Speaker Verification Fairness Adversarial Reweighting

Adversarial Reweighting

Papers

Improving Accuracy-robustness Trade-off via Pixel Reweighted Adversarial Training

Adversarial Reweighting with $\alpha$-Power Maximization for Domain Adaptation

Adversarial Reweighting Guided by Wasserstein Distance for Bias Mitigation

Doubly Robust Instance-Reweighted Adversarial Training

Adversarial Reweighting for Speaker Verification Fairness