As machine learning (ML) systems are increasingly adopted in high-stakes decision-making domains, ensuring fairness in their outputs has become a central challenge. At the core of fair ML research are the datasets used to investigate bias and develop mitigation strategies. Yet, much of the existing work relies on a narrow selection of datasets--often arbitrarily chosen, inconsistently processed, and lacking in diversity--undermining the generalizability and reproducibility of results. To address these limitations, we present FairGround: a unified framework, data corpus, and Python package aimed at advancing reproducible research and critical data studies in fair ML classification. FairGround currently comprises 44 tabular datasets, each annotated with rich fairness-relevant metadata. Our accompanying Python package standardizes dataset loading, preprocessing, transformation, and splitting, streamlining experimental workflows. By providing a diverse and well-documented dataset corpus along with robust tooling, FairGround enables the development of fairer, more reliable, and more reproducible ML models. All resources are publicly available to support open and collaborative research.

Differential privacy (DP) is a prominent method for protecting information about individuals during data analysis. Training neural networks with differentially private stochastic gradient descent (DPSGD) influences the model's learning dynamics and, consequently, its output. This can affect the model's performance and fairness. While the majority of studies on the topic report a negative impact on fairness, it has recently been suggested that fairness levels comparable to non-private models can be achieved by optimizing hyperparameters for performance directly on differentially private models (rather than re-using hyperparameters from non-private models, as is common practice). In this work, we analyze the generalizabil-ity of this claim by 1) comparing the disparate impact of DPSGD on different performance metrics, and 2) analyzing it over a wide range of hyperparameter settings. We highlight that a disparate impact on one metric does not necessarily imply a disparate impact on another. Most importantly, we show that while optimizing hyperparameters directly on differentially private models does not mitigate the disparate impact of DPSGD reliably, it can still lead to improved utility-fairness trade-offs compared to re-using hyperparameters from non-private models. We stress, however, that any form of hyperparameter tuning entails additional privacy leakage, calling for careful considerations of how to balance privacy, utility and fairness. Finally, we extend our analyses to DPSGD-Global-Adapt, a variant of DPSGD designed to mitigate the disparate impact on accuracy, and conclude that this alternative may not be a robust solution with respect to hyperparameter choice.

While researchers have been trying to understand the stress and fatigue among pilots, especially pilot trainees, and to develop stress/fatigue models to automate the process of detecting stress/fatigue, they often do not consider biases such as sex in those models. However, in a critical profession like aviation, where the demographic distribution is disproportionately skewed to one sex, it is urgent to mitigate biases for fair and safe model predictions. In this work, we investigate the perceived stress/fatigue of 69 college students, including 40 pilot trainees with around 63% male. We construct models with decision trees first without bias mitigation and then with bias mitigation using a threshold optimizer with demographic parity and equalized odds constraints 30 times with random instances. Using bias mitigation, we achieve improvements of 88.31% (demographic parity difference) and 54.26% (equalized odds difference), which are also found to be statistically significant.

In the healthcare industry, researchers have been developing machine learning models to automate diagnosing patients with respiratory illnesses based on their breathing patterns. However, these models do not consider the demographic biases, particularly sex bias, that often occur when models are trained with a skewed patient dataset. Hence, it is essential in such an important industry to reduce this bias so that models can make fair diagnoses. In this work, we examine the bias in models used to detect breathing patterns of two major respiratory diseases, i.e., chronic obstructive pulmonary disease (COPD) and COVID-19. Using decision tree models trained with audio recordings of breathing patterns obtained from two open-source datasets consisting of 29 COPD and 680 COVID-19-positive patients, we analyze the effect of sex bias on the models. With a threshold optimizer and two constraints (demographic parity and equalized odds) to mitigate the bias, we witness 81.43% (demographic parity difference) and 71.81% (equalized odds difference) improvements. These findings are statistically significant.

Low-rank adaptation of large models, particularly LoRA, has gained traction due to its computational efficiency. This efficiency, contrasted with the prohibitive costs of full-model fine-tuning, means that practitioners often turn to LoRA and sometimes without a complete understanding of its ramifications. In this study, we focus on fairness and ask whether LoRA has an unexamined impact on utility, calibration, and resistance to membership inference across different subgroups (e.g., genders, races, religions) compared to a full-model fine-tuning baseline. We present extensive experiments across vision and language domains and across classification and generation tasks using ViT-Base, Swin-v2-Large, Llama-2 7B, and Mistral 7B. Intriguingly, experiments suggest that while one can isolate cases where LoRA exacerbates model bias across subgroups, the pattern is inconsistent -- in many cases, LoRA has equivalent or even improved fairness compared to the base model or its full fine-tuning baseline. We also examine the complications of evaluating fine-tuning fairness relating to task design and model token bias, calling for more careful fairness evaluations in future work.

Model-induced distribution shifts (MIDS) occur as previous model outputs pollute new model training sets over generations of models. This is known as model collapse in the case of generative models, and performative prediction or unfairness feedback loops for supervised models. When a model induces a distribution shift, it also encodes its mistakes, biases, and unfairnesses into the ground truth of its data ecosystem. We introduce a framework that allows us to track multiple MIDS over many generations, finding that they can lead to loss in performance, fairness, and minoritized group representation, even in initially unbiased datasets. Despite these negative consequences, we identify how models might be used for positive, intentional, interventions in their data ecosystems, providing redress for historical discrimination through a framework called algorithmic reparation (AR). We simulate AR interventions by curating representative training batches for stochastic gradient descent to demonstrate how AR can improve upon the unfairnesses of models and data ecosystems subject to other MIDS. Our work takes an important step towards identifying, mitigating, and taking accountability for the unfair feedback loops enabled by the idea that ML systems are inherently neutral and objective.