Skip to main content

Diagnosing, Understanding, and Fixing Data Biases for Trusted Data Science

Romila PradhanPI: Romila Pradhan

Assistant Professor in Computer and Information Technology
Purdue University


Data iconFramework component: Data

The thesis of this proposal is that data preparation tasks tailored to downstream machine learning (ML) applications can serve as a basis for detecting and mitigating algorithmic bias. The overarching goal of this project is to demonstrate value for the importance of data quality in establishing public trust in data-driven decision making, in part by tracing bias in ML models and pipelines. The project will investigate how to diagnose bias, as well as how to evaluate and integrate the impact of data quality on bias in downstream ML tasks. By decoupling data-based applications from the mechanics of managing data quality, the project will make it easier for practitioners to detect and mitigate biases stemming from data throughout their workflows.

Outcomes and Updates
Back to top