Computer Science Colloquium: Jinzhen Wang - Toward smart and efficient data management

Speaker

Jinzhen Wang, Assistant Professor (Brooklyn College of CUNY)

Abstract

Extreme-scale simulations running on leadership-class supercomputers are crucial for scientific advancements in various domains. Yet, the relentless pursuit of higher resolution and fidelity has led to an exponential surge in the volume of scientific data stemming from these simulations. Consequently, the formidable tasks of storage and analysis have emerged as pressing challenges.

To address this challenge, lossy compression has emerged to reduce the storage footprint. However, a critical concern revolving around the accuracy preservation of lossy compression, especially for downstream analyses has impeded the widespread integration of lossy compression into the high-performance computing workflow.

In this talk, I will introduce my Ph.D. research toward smart and efficient scientific data management. For efficient data management, it is crucial to achieve the most benefit from lossy data compression. I will introduce our work on performance modeling of data compression, utilizing both analytic and deep learning approaches.

In pursuit of smart data management, we aim to determine whether we can achieve better data compression with deep learning approaches. I will present our work on 1) scientific data compression with Autoencoder network and 2) minimizing the I/O overhead of scientific data retrieval using deep neural networks.

Finally, I will conclude by outlining potential research directions and future work aimed at fostering smarter and more efficient HPC scientific data management.

Speaker Biography

Jinzhen Wang is an Assistant Professor in the Department of Computer and Information Science at Brooklyn College of CUNY. He received his Ph.D. in Electrical Engineering from the New Jersey Institute of Technology in 2023. Previously, he held a research internship at the Los Alamos National Laboratory. His research focuses on high-performance computing, with the emphasis on data compression, data management and machine learning.

Review our Building Entry Policy for in-person events.