Multi-task learning (MTL) is a paradigm that aims to improve the generalization of models by simultaneously learning multiple related tasks, leveraging shared representations and task-specific information to enhance performance on individual tasks. However, existing work has shown that MTL can potentially hinder generalization, with one key factor being spurious correlations between tasks. Owing to the knowledge-sharing property, the per-task predictors are more likely to develop reliance on spurious features. Most existing approaches address this issue through distributional robustness, aiming to maintain consistent performance across different distributions under unknown covariate shifts. However, this formulation lacks theoretical guarantees and can be sensitive to the construction of covariate shifts. In this work, we propose a novel perspective, where we seek to identify spurious correlations between tasks. Drawing inspirations from conventional formulations on spurious correlation, for each task, we propose to distinguish its spurious tasks using the difference in correlation coefficients between the empirical distribution and class-wise resampled distributions, thereby capturing the correlations between task labels w.r.t. each class. We prove theoretically the feasibility of the resampling strategy in characterizing spurious correlations between tasks. Furthermore, we propose a simple fine-tuning strategy, de-biased adversarial training, where the per-task predictors are adversarially trained to disregard information associated with their spurious tasks. Experimental results on six benchmark datasets show that our method effectively mitigates spurious correlations and outperforms state-of-the-art methods in improving generalization.

Achieving fairness in machine learning remains a critical challenge, especially due to the opaque effects of data augmentation on input spaces within nonlinear neural networks. Nevertheless, current approaches that emphasize augmenting latent features, rather than input spaces, offer limited insights into their ability to detect and mitigate bias. In response, we introduce the concept of the "unfair region" in the latent space, a subspace that highlights areas where misclassification rates for certain demographic groups are disproportionately high, leading to unfair prediction results. To address this, we propose Adversarial Latent Feature Augmentation (ALFA), a method that leverages adversarial fairness attacks to perturb latent space features, which are then used as data augmentation for fine-tuning. ALFA intentionally shifts latent features into unfair regions, and the last layer of the network is fine-tuned with these perturbed features, leading to a corrected decision boundary that enhances fairness in classification in a cost-effective manner. We present a theoretical framework demonstrating that our adversarial fairness objective reliably generates biased feature perturbations, and that fine-tuning on samples from these unfair regions ensures fairness improvements. Extensive experiments across diverse datasets, modalities, and backbone networks validate that training with these adversarial features significantly enhances fairness while maintaining predictive accuracy in classification tasks.

Multimodal models or Vision Language Models (VLMs) have reshaped the paradigm in machine learning, offering zero-shot capabilities that require no additional training when adapted to new classification tasks. However, despite their advancements, spurious correlations still exist in VLMs. Existing approaches to tackle this issue often require target label annotations, contradicting the principle of zero-shot classification, or they primarily focus on a single modality, risking misalignment between text and image modalities. Others rely on extensive domain knowledge or large language models (LLMs) to characterize spurious features, making the performance sensitive to the generated prompts and undermining zero-shot capability. In response, we propose a new solution that tackles spurious correlations in VLMs within the zero-shot setting. Our approach utilizes a translation operation that preserves the latent space distribution to address issues of spurious correlations. In particular, our method is grounded in and inspired by a theoretical analysis, which identifies that the optimal translation directions are along the spurious vector. As VLMs unify two modalities, we compute spurious vectors from the text prompts and guide the translation for image embeddings, aligning the requirements for the fusion of different modalities in VLMs. We conducted experiments on benchmark datasets, which have shown significant improvements in worst-group accuracy. Additionally, our visualizations of VLMs further demonstrate the effectiveness of this intervention.

While numerous work has been proposed to address fairness in machine learning, existing methods do not guarantee fair predictions under imperceptible feature perturbation, and a seemingly fair model can suffer from large group-wise disparities under such perturbation. Moreover, while adversarial training has been shown to be reliable in improving a model's robustness to defend against adversarial feature perturbation that deteriorates accuracy, it has not been properly studied in the context of adversarial perturbation against fairness. To tackle these challenges, in this paper, we study the problem of adversarial attack and adversarial robustness w.r.t. two terms: fairness and accuracy. From the adversarial attack perspective, we propose a unified structure for adversarial attacks against fairness which brings together common notions in group fairness, and we theoretically prove the equivalence of adversarial attacks against different fairness notions. Further, we derive the connections between adversarial attacks against fairness and those against accuracy. From the adversarial robustness perspective, we theoretically align robustness to adversarial attacks against fairness and accuracy, where robustness w.r.t. one term enhances robustness w.r.t. the other term. Our study suggests a novel way to unify adversarial training w.r.t. fairness and accuracy, and experiments show our proposed method achieves better robustness w.r.t. both terms.

Although machine learning algorithms demonstrate impressive performance, their trustworthiness remains a critical issue, particularly concerning fairness when implemented in real-world applications. Many notions of group fairness aim to minimize disparities in performance across protected groups. However, it can inadvertently reduce performance in certain groups, leading to sub-optimal outcomes. In contrast, Min-max group fairness notion prioritizes the improvement for the worst-performing group, thereby advocating a utility-promoting approach to fairness. However, it has been proven that existing efforts to achieve Min-max fairness exhibit limited effectiveness. In response to this challenge, we leverage the recently proposed "Neural Collapse'' framework to re-examine Empirical Risk Minimization (ERM) training, specifically investigating the root causes of poor performance in minority groups. The layer-peeled model is employed to decompose a network into two parts: an encoder to learn latent representation, and a subsequent classifier, with a systematic characterization of their training behaviors being conducted. Our analysis reveals that while classifiers achieve maximum separation, the separability of representations is insufficient, particularly for minority groups. This indicates the sub-optimal performance in minority groups stems from less separable representations, rather than classifiers. To tackle this issue, we introduce a novel strategy that incorporates a frozen classifier to directly enhance representation. Furthermore, we introduce two easily implemented loss functions to guide the learning process. The experimental assessments carried out on real-world benchmark datasets spanning the domains of Computer Vision, Natural Language Processing, and Tabular data demonstrate that our approach outperforms existing state-of-the-art methods in promoting the Min-max fairness notion.

Fairness is now an important issue in machine learning. There are arising concerns that automated decision-making systems reflect real-world biases. Although a wide range of fairness-related methods have been proposed in recent years, the under-representation problem has been less studied. Due to the uneven distribution of samples from different populations, machine learning models tend to be biased against minority groups when trained by minimizing the average empirical risk across all samples. In this paper, we propose a novel adaptive reweighing method to address representation bias. The goal of our method is to achieve group-level balance among different demographic groups by learning adaptive weights for each sample. Our approach emphasizes more on error-prone samples in prediction and enhances adequate representation of minority groups for fairness. We derive a closed-form solution for adaptive weight assignment and propose an efficient algorithm with theoretical convergence guarantees. We theoretically analyze the fairness of our model and empirically verify that our method strikes a balance between fairness and accuracy. In experiments, our method achieves comparable or better performance than state-of-the-art methods in both classification and regression tasks. Furthermore, our method exhibits robustness to label noise on various benchmark datasets.

Most of existing work on fairness assumes available demographic information in the training set. In practice, due to legal or privacy concerns, when demographic information is not available in the training set, it is crucial to find alternative objectives to ensure fairness. Existing work on fairness without demographics follows Rawlsian Max-Min fairness objectives. However, such constraints could be too strict to improve group fairness, and could lead to a great decrease in accuracy. In light of these limitations, in this paper, we propose to solve the problem from a new perspective, i.e., through knowledge distillation. Our method uses soft label from an overfitted teacher model as an alternative, and we show from preliminary experiments that soft labelling is beneficial for improving fairness. We analyze theoretically the fairness of our method, and we show that our method can be treated as an error-based reweighing. Experimental results on three datasets show that our method outperforms state-of-the-art alternatives, with notable improvements in group fairness and with relatively small decrease in accuracy.

Fairness has become an important topic in machine learning. Generally, most literature on fairness assumes that the sensitive information, such as gender or race, is present in the training set, and uses this information to mitigate bias. However, due to practical concerns like privacy and regulation, applications of these methods are restricted. Also, although much of the literature studies supervised learning, in many real-world scenarios, we want to utilize the large unlabelled dataset to improve the model's accuracy. Can we improve fair classification without sensitive information and without labels? To tackle the problem, in this paper, we propose a novel reweighing-based contrastive learning method. The goal of our method is to learn a generally fair representation without observing sensitive attributes.Our method assigns weights to training samples per iteration based on their gradient directions relative to the validation samples such that the average top-k validation loss is minimized. Compared with past fairness methods without demographics, our method is built on fully unsupervised training data and requires only a small labelled validation set. We provide rigorous theoretical proof of the convergence of our model. Experimental results show that our proposed method achieves better or comparable performance than state-of-the-art methods on three datasets in terms of accuracy and several fairness metrics.