12/25/2023
By Seung Woo Son
The Francis College of Engineering, Department of Electrical and Computer Engineering, cordially invites you to a dissertation proposal by Jiaxi Chen on "Optimization of DCT-based Lossy Compression Algorithm for Scientific Datasets."
Candidate Name: Jiaxi Chen
Date: Tuesday, Jan. 9, 2024
Time: 2 to 4 p.m. EST
Location: Ball 302 and via Zoom. Those interested in attending via Zoom should contact the Ph.D. advisor (seungwoo_son@uml.edu) at least 24 hours before the proposal defense to request.
Committee Members:
- Committee Chair, Advisor, Seung Woo Son, Associate Professor of Electrical and Computer Engineering, UMass Lowell
- Xuejun Lu, Professor of Electrical and Computer Engineering, UMass Lowell
- Yan Luo, Professor of Electrical and Computer Engineering, UMass Lowell
Abstract:
High-performance computing (HPC) systems that run scientific simulations of significance produce a large amount of data during runtime. Transferring or storing such big datasets causes a severe I/O bottleneck and a considerable storage burden. One approach to mitigate such overhead is to reduce data size by compression techniques before transferring it through the network or storing it on disk. Traditional lossless compression can preserve all the information with full fidelity but cannot achieve appreciable data reduction. Unlike lossless compression algorithms, error-controlled lossy compressors could significantly reduce the data size while respecting the user-defined error bound.
This dissertation focuses on optimizing DCTZ, one of the transform-based lossy compressors with a highly efficient encoding and purpose-built error control mechanism that accomplishes high compression ratios with high data fidelity. DCTZ reduces the input data size by preserving only small parts of coefficients in the frequency domain. However, since DCTZ quantizes the DCT coefficients in the frequency domain, it may only partially control the relative error bound defined by the user. This dissertation investigates the performance of DCTZ with real-world scientific datasets, improves the compression quality, and proposes a new mode of DCTZ.
Specifically, three techniques have been proposed to optimize improving compression ratios and lowering reconstruction errors. The first is based on the current implementation of DCTZ, where we propose a preconditioning method based on level offsetting and scaling to control the magnitude of input of the DCTZ framework, thereby enforcing stricter error bounds. The second is predicting the compression ratio of three DCT-based compression frameworks using multiple machine-learning models. The third is implementing the fixed-PSNR mode of DCTZ, which allows the user to define the desired PSNR before compression. We will also present the evaluation of our framework with other state-of-the-art lossy compressors.
For more information, please get in touch with Jiaxi Chen.