Data Attribution: A Principled Approach for Data-Centric AI

Location: BBB 3901
Zoom: https://umich.zoom.us/j/97981954169
Meeting ID: 979 8195 4169
Passcode: aiseminar

Abstract

Data plays an increasingly crucial role in both the performance and the safety of AI models. Data attribution is an emerging family of techniques aimed at quantifying the impact of individual training data points on a model trained on them, which has found data-centric applications such as training data curation, instance-based explanation, and data markets. In this talk, I will present our recent work on data attribution, covering both applications and methodology. On the application side, I will discuss its use in addressing generative AI copyright challenges and in enabling robust counterfactual explanations under dataset shift. On the methodology side, I will introduce novel methods that extend data attribution’s applicability to broader machine learning paradigms, as well as that improve the efficacy-efficiency tradeoffs.

Bio

Jiaqi Ma is an Assistant Professor at the University of Illinois Urbana-Champaign (UIUC). His research interests lie in the broad area of trustworthy and data-centric AI, with recent focuses including data attribution, training data curation, explainable AI, and copyright issues of generative AI. Jiaqi’s work has been recognized with the Gary M. Olson Outstanding Student Award from University of Michigan, a Best Paper Award form the DPFM Workshop at ICLR 2024, and New Faculty Highlight at AAAI 2025. Prior to joining UIUC, Jiaqi earned his PhD from the University of Michigan and worked as a postdoctoral researcher at Harvard University.