Ozan TUĞLUK, Scientific Computing Program/IAM, “Whither data? Scientific computing and compression in the exascale era.”

Abstract: Rapidly evolving high performance computing resources result in an exponential increase in data storage demands. Scientific data compression is a much cheaper alternative to constantly expanding data storage units. However, due to the nature of current floating point representations and scientific simulations, lossless compression is rarely relevant. Lossy compression offers a viable alternative, however there are two main challenges, namely hesitancy of applications scientist and seemingly low compression ratios. We present our approach to tackling these two problems. Our approach involves compressing the data so that a prescribed loss is guaranteed, not on the original data, but on the quantities of interest which are computed from the data. Advantages of this approach are twofold, preserving quantities of interest results in better compression, and assures the user that compression artifacts will not adversely effect their analysis. We present our mathematical framework, and our compression tool/library MGARD (MultiGrid Adaptive Reduction of Data) which was developed for CODAR (Co-design Center for Online Data Analysis and Reduction at the Exascale). We provide illustrative examples from experimental and computational data.

Date: December 12, 2025

Time: 14:30-15:30

Venue: Institute of Applied Mathematics (IAM), Hayri Körezlioğlu Seminar Room (S-212)