Xuran Li

Xuran Li, Peng Wu, Jing Su

Accuracy and individual fairness are both crucial for trustworthy machine learning, but these two aspects are often incompatible with each other so that enhancing one aspect may sacrifice the other inevitably with side effects of true bias or false fairness. We propose in this paper a new fairness criterion, accurate fairness, to align individual fairness with accuracy. Informally, it requires the treatments of an individual and the individual's similar counterparts to conform to a uniform target, i.e., the ground truth of the individual. We prove that accurate fairness also implies typical group fairness criteria over a union of similar sub-populations. We then present a Siamese fairness in-processing approach to minimize the accuracy and fairness losses of a machine learning model under the accurate fairness constraints. To the best of our knowledge, this is the first time that a Siamese approach is adapted for bias mitigation. We also propose fairness confusion matrix-based metrics, fair-precision, fair-recall, and fair-F1 score, to quantify a trade-off between accuracy and individual fairness. Comparative case studies with popular fairness datasets show that our Siamese fairness approach can achieve on average 1.02%-8.78% higher individual fairness (in terms of fairness through awareness) and 8.38%-13.69% higher accuracy, as well as 10.09%-20.57% higher true fair rate, and 5.43%-10.01% higher fair-F1 score, than the state-of-the-art bias mitigation techniques. This demonstrates that our Siamese fairness approach can indeed improve individual fairness without trading accuracy. Finally, the accurate fairness criterion and Siamese fairness approach are applied to mitigate the possible service discrimination with a real Ctrip dataset, by on average fairly serving 112.33% more customers (specifically, 81.29% more customers in an accurately fair way) than baseline models.

Xuran Li, Jingyi Wang

Large language models (LLMs) are increasingly deployed in real-world systems, yet they can produce toxic or biased outputs that undermine safety and trust. Post-hoc model repair provides a practical remedy, but the high cost of parameter updates motivates selective use of repair data. Despite extensive prior work on data selection for model training, it remains unclear which sampling criteria are most effective and efficient when applied specifically to behavioral repair of large generative models. Our study presents a systematic analysis of sample prioritization strategies for LLM repair. We evaluate five representative selection methods, including random sampling, K-Center, gradient-norm-based selection(GraNd), stratified coverage (CCS), and a Semantic-Aware Prioritized Sampling (SAPS) approach we proposed. Repair effectiveness and trade-offs are assessed through toxicity reduction, perplexity on WikiText-2 and LAMBADA, and three composite metrics: the Repair Proximity Score (RPS), the Overall Performance Score (OPS), and the Repair Efficiency Score (RES). Experimental results show that SAPS achieves the best balance between detoxification, utility preservation, and efficiency, delivering comparable or superior repair outcomes with substantially less data. Random sampling remains effective for large or robust models, while high-overhead methods such as CCS and GraNd provide limited benefit. The optimal data proportion depends on model scale and repair method, indicating that sample selection should be regarded as a tunable component of repair pipelines. Overall, these findings establish selection-based repair as an efficient and scalable paradigm for maintaining LLM reliability.

Xuran Li, Jingyi Wang, Xiaohan Yuan et al.

It is often desirable to remove (a.k.a. unlearn) a specific part of the training data from a trained neural network model. A typical application scenario is to protect the data holder's right to be forgotten, which has been promoted by many recent regulation rules. Existing unlearning methods involve training alternative models with remaining data, which may be costly and challenging to verify from the data holder or a thirdparty auditor's perspective. In this work, we provide a new angle and propose a novel unlearning approach by imposing carefully crafted "patch" on the original neural network to achieve targeted "forgetting" of the requested data to delete. Specifically, inspired by the research line of neural network repair, we propose to strategically seek a lightweight minimum "patch" for unlearning a given data point with certifiable guarantee. Furthermore, to unlearn a considerable amount of data points (or an entire class), we propose to iteratively select a small subset of representative data points to unlearn, which achieves the effect of unlearning the whole set. Extensive experiments on multiple categorical datasets demonstrates our approach's effectiveness, achieving measurable unlearning while preserving the model's performance and being competitive in efficiency and memory consumption compared to various baseline methods.

Xuran Li, Peng Wu, Yanting Chen et al.

Deep neural networks (DNNs) are known to be sensitive to adversarial input perturbations, leading to a reduction in either prediction accuracy or individual fairness. To jointly characterize the susceptibility of prediction accuracy and individual fairness to adversarial perturbations, we introduce a novel robustness definition termed robust accurate fairness. Informally, robust accurate fairness requires that predictions for an instance and its similar counterparts consistently align with the ground truth when subjected to input perturbations. We propose an adversarial attack approach dubbed RAFair to expose false or biased adversarial defects in DNN, which either deceive accuracy or compromise individual fairness. Then, we show that such adversarial instances can be effectively addressed by carefully designed benign perturbations, correcting their predictions to be accurate and fair. Our work explores the double-edged sword of input perturbations to robust accurate fairness in DNN and the potential of using benign perturbations to correct adversarial instances.

Xuran Li, Peng Wu, Kaixiang Dong et al.

Deep neural networks (DNNs) often face challenges due to their vulnerability to various adversarial perturbations, including false perturbations that undermine prediction accuracy and biased perturbations that cause biased predictions for similar inputs. This paper introduces a novel approach, RobustFair, to evaluate the accurate fairness of DNNs when subjected to these false or biased perturbations. RobustFair employs the notion of the fairness confusion matrix induced in accurate fairness to identify the crucial input features for perturbations. This matrix categorizes predictions as true fair, true biased, false fair, and false biased, and the perturbations guided by it can produce a dual impact on instances and their similar counterparts to either undermine prediction accuracy (robustness) or cause biased predictions (individual fairness). RobustFair then infers the ground truth of these generated adversarial instances based on their loss function values approximated by the total derivative. To leverage the generated instances for trustworthiness improvement, RobustFair further proposes a data augmentation strategy to prioritize adversarial instances resembling the original training set, for data augmentation and model retraining. Notably, RobustFair excels at detecting intertwined issues of robustness and individual fairness, which are frequently overlooked in standard robustness and individual fairness evaluations. This capability empowers RobustFair to enhance both robustness and individual fairness evaluations by concurrently identifying defects in either domain. Empirical case studies and quantile regression analyses on benchmark datasets demonstrate the effectiveness of the fairness confusion matrix guided perturbation for false or biased adversarial instance generation.