Beepul Bharti

preprints

publications

1. arXiv

   Global Sequential Testing for Multi-Stream Auditing

   Bharti, B., Pal, A., and Sulam, J.

   Abstract

   Across many risk-sensitive areas, it is critical to continuously audit the performance of machine learning systems and detect any unusual behavior quickly. This can be modeled as a sequential hypothesis testing problem with k incoming streams of data and a global null hypothesis that asserts that the system is working as expected across all k streams. The standard global test employs a Bonferroni correction and has an expected stopping time bound of O(ln(k/α)) when k is large and the significance level of the test, α, is small. In this work, we construct new sequential tests by using ideas of merging test martingales with different trade-offs in expected stopping times under different, sparse or dense alternative hypotheses. We further derive a new, balanced test that achieves an improved expected stopping time bound that matches Bonferroni’s in the sparse setting but that naturally results in O(1/k ln(1/α)) under a dense alternative. We empirically demonstrate the effectiveness of our proposed tests on synthetic and real-world data.

2. Rad

   Pitfalls and Best Practices in Evaluation of AI Algorithmic Biases in Radiology

   Yi, P., Bachina, P.,  Bharti, B., Garin, S., Kanhere, A., Kulkarni, P., Li, D., Parekh, V., Santomartino, S., Moy, L., and others, .

   Abstract

   Despite growing awareness of problems with fairness in artificial intelligence (AI) models in radiology, evaluation of algorithmic biases, or AI biases, remains challenging due to various complexities. These include incomplete reporting of demographic information in medical imaging datasets, variability in definitions of demographic categories, and inconsistent statistical definitions of bias. To guide the appropriate evaluation of AI biases in radiology, this article summarizes the pitfalls in the evaluation and measurement of algorithmic biases. These pitfalls span the spectrum from the technical (eg, how different statistical definitions of bias impact conclusions about whether an AI model is biased) to those associated with social context (eg, how different conventions of race and ethnicity impact identification or masking of biases). Actionable best practices and future directions to avoid these pitfalls are summarized across three key areas: (a) medical imaging datasets, (b) demographic definitions, and (c) statistical evaluations of bias. Although AI bias in radiology has been broadly reviewed in the recent literature, this article focuses specifically on underrecognized potential pitfalls related to the three key areas. By providing awareness of these pitfalls along with actionable practices to avoid them, exciting AI technologies can be used in radiology for the good of all people.

3. May. 01, 2025

   ICML

   Multiaccuracy and Multicalibration via Proxy Groups

   Bharti, B., Clemens-Sewall, M., Yi, P., and Sulam, J.

   Abstract

   As the use of predictive machine learning algorithms increases in high-stakes decision-making, it is imperative that these algorithms are fair across sensitive groups. However, measuring and enforcing fairness in real-world applications can be challenging due to missing or incomplete sensitive group data. Proxy-sensitive attributes have been proposed as a practical and effective solution in these settings, but only for parity-based fairness notions. Knowing how to evaluate and control for fairness with missing sensitive group data for newer and more flexible frameworks, such as multiaccuracy and multicalibration, remains unexplored. In this work, we address this gap by demonstrating that in the absence of sensitive group data, proxy-sensitive attributes can provably be used to derive actionable upper bounds on the true multiaccuracy and multicalibration, providing insights into a model’s potential worst-case fairness violations. Additionally, we show that adjusting models to satisfy multiaccuracy and multicalibration across proxy-sensitive attributes can significantly mitigate these violations for the true, but unknown, demographic groups. Through several experiments on real-world datasets, we illustrate that approximate multiaccuracy and multicalibration can be achieved even when sensitive group data is incomplete or unavailable.

4. March. 12, 2025

   CPAL

   Sufficient and Necessary Explanations (and What Lies in Between)

   Bharti, B., Yi, P., and Sulam, J.

   Abstract

   As complex machine learning models continue to be used in high-stakes decision settings, understanding their predictions is crucial. Post-hoc explanation methods aim to identify which features of an input x are important to a model’s prediction f(x). However, explanations often vary between methods and lack clarity, limiting the information we can draw from them. To address this, we formalize two precise concepts—sufficiency and necessity—to quantify how features contribute to a model’s prediction. We demonstrate that, although intuitive and simple, these two types of explanations may fail to fully reveal which features a model deems important. To overcome this, we propose and study a unified notion of importance that spans the entire sufficiency-necessity axis. Our unified notion, we show, has strong ties to notions of importance based on conditional independence and Shapley values. Lastly, through various experiments, we quantify the sufficiency and necessity of popular post-hoc explanation methods. Furthermore, we show that generating explanations along the sufficiency-necessity axis can uncover important features that may otherwise be missed, providing new insights into feature importance.

5. Dec. 13, 2023

   NeurIPS

   Estimating and Controlling for Equalized Odds via Sensitive Attribute Predictors

   Bharti, B., Yi, P., and Sulam, J.

   Abstract

   As the use of machine learning models in real world high-stakes decision settings continues to grow, it is highly important that we are able to audit and control for any potential fairness violations these models may exhibit towards certain groups. To do so, one naturally requires access to sensitive attributes, such as demographics, biological sex, or other potentially sensitive features that determine group membership. Unfortunately, in many settings, this information is often unavailable. In this work we study the well known equalized odds (EOD) definition of fairness. In a setting without sensitive attributes, we first provide tight and computable upper bounds for the EOD violation of a predictor. These bounds precisely reflect the worst possible EOD violation. Second, we demonstrate how one can provably control the worst-case EOD by a new post-processing correction method. Our results characterize when directly controlling for EOD with respect to the predicted sensitive attributes is – and when is not – optimal when it comes to controlling worst-case EOD. Our results hold under assumptions that are milder than previous works, and we illustrate these results with experiments on synthetic and real datasets.

6. Dec. 12, 2023

   TMLR

   SHAP-XRT: The Shapley Value Meets Conditional Independence Testing

   Teneggi, J.*,  Bharti, B.*, Romano, Y., and Sulam, J.

   Abstract

   The complex nature of artificial neural networks raises concerns on their reliability, trustworthiness, and fairness in real-world scenarios. The Shapley value—a solution concept from game theory—is one of the most popular explanation methods for machine learning models. More traditionally, from a statistical perspective, feature importance is defined in terms of conditional independence. So far, these two approaches to interpretability and feature importance have been considered separate and distinct. In this work, we show that Shapley-based explanation methods and conditional independence testing are closely related. We introduce the SHAPley-EXplanation Randomization Test (SHAP-XRT), a testing procedure inspired by the Conditional Randomization Test (CRT) for a specific notion of local (i.e., on a sample) conditional independence. With it, we prove that for binary classification problems, the marginal contributions in the Shapley value provide lower and upper bounds to the p-values of their respective tests. Furthermore, we show that the Shapley value itself provides an upper bound to the p-value of a global (i.e., overall) null hypothesis. As a result, we further our understanding of Shapley-based explanation methods from a novel perspective and characterize under which conditions one can make statistically valid claims about feature importance via the Shapley value.

7. Aug. 30, 2022

   CAR

   Sex imbalance produces biased deep learning models for knee osteoarthritis detection

   Li, D.,  Bharti, B., Wei, J., Sulam, J., and Yi, P.