Learning From Aggregated Responses: Improving Model Utility Under Privacy Constraints

Abstract:
In many real-world scenarios, training data is aggregated before being shared with the learner to protect users’ privacy. This talk explores recent advancements in aggregate learning, where datasets are grouped into bags of samples with only summary responses available for each bag. I will discuss novel loss construction and bagging schemes that enhance model accuracy while maintaining privacy. Key topics include using priors to inform bag construction and an iterative boosting algorithm that refines priors through sample splitting. Additionally, I will present learning rules in this framework that achieve PAC learning guarantees for classification. While Empirical Proportional Risk Minimization (EPRM) achieves fast rates under realizability, it can falter in agnostic settings. To address this limitation, I will introduce a debiased proportional square loss that achieves “optimistic rates” in both realizable and agnostic scenarios.

Bio:
Adel Javanmard is a Professor in the department of Data Sciences and Operation, at USC Marshall School of Business, and an executive member of the iORB committee. Previously he was the Dean’s Associate chair. He received his PhD in Electrical Engineering from Stanford University in 2014. His research interests are broadly in the area of high-dimensional statistics, machine learning, optimization, and personal decision making. Adel is the recipient of several highly prestigious awards, including the Alfred Sloan research fellow in mathematics, IMS Tweedie New Researcher Award, NSF CAREER award, Faculty Research Awards from Amazon, Adobe, Google and the Thomas Cover dissertation award from IEEE Society. He was the recipient of the Dr. Douglas Basil Award for Junior Business Faculty (2018) and the Deans’ award for Research Excellence (2024). He also received the USC Golden Apple Award for teaching in undergraduate program (2022)