Improving Quality of Lossy Compression by Feature Regeneration

This project is generally aimed to contribute to the research and development of lossy compression frameworks focusing on improving the compression quality for scientific applications by considering feature regeneration. This subcontract will support a graduate student who is supervised by Professor Xin Liang at University of Kentucky to work on this project virtually. Scope This project (subcontract) will focus on improving the compression quality of scientific lossy data compressors by exploring intelligent methods and parameter optimization strategies to regenerate feature and mitigate artifacts. The project will also focus on optimizing the performance and quality of the underlying methods for ECP applications. Objectives The goal of this subcontract is to i) study new techniques to regenerate features and/or mitigate artifacts for scientific lossy data compressors; ii) integrate the methods to improve the compression quality of SZ using parameter optimizations; iii) evaluate and optimize the framework with ECP applications.