April 24, 2025
12:15 PM – 1:15 PM
Speaker:
Xin Liang, Computer Science Department, University of Kentucky
Where:
327 McVey Hall
(Zoom link: https://uky.zoom.us/j/82467171189)
Title:
Advancing Extreme-scale Data Science via Trust-driven Data Reduction
Abstract:
Extreme-scale scientific simulations and experiments generate more data than that can be stored, transmitted, and analyzed. The recently delivered exascale systems and high-resolution scientific instruments are exacerbating this problem, due to the imbalanced growth between storage systems and data. My research focuses on the development of efficient algorithms and scalable software for high-performance data management on massively parallel and heterogeneous systems. In this talk, I will present how we address the scientific data challenges arising from real-world scientific applications through trust- driven data reduction. Specifically, I will talk about 1) error-controlled and feature- preserving lossy data reduction; 2) data refactoring and error-controlled progressive retrieval; and 3) throughput improvement with accelerators. Our ultimate goal is to deliver an intelligent data management system with trust-driven data reduction that enables flexible data storage, on-demand data retrieval, and adaptive data analytics, which could greatly reduce the time to obtain insights from data and thus advance scientific discoveries.