Evaluating the Performance Acceleration of Generalized Linear Solver using Normal Equation on Three Architectures for Tall Skinny Datasets

• Published in: 2022 International Conference on Computational Science and Computational Intelligence (CSCI) pp. 134 - 139
• Main Authors: Van Sang, Tran, Yamaguchi, Rie Shigetomi, Kobayashi, Ryosuke, Nakata, Toshiyuki
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-12-2022
• Subjects: compilers; Computational modeling; Computer architecture; generalized linear model; Mathematical models; normal equation; Scalability; Scientific computing; Support vector machines; tall skinny data; Training; vector machine
• ISSN: 2769-5654
• DOI: 10.1109/CSCI58124.2022.00028

In previous work, we effectively applied a Normal Equation method to solve the most draining task in the Generalized Linear Model training process on a tall-skinny real-world dataset. This paper generalizes this method by applying it to synthetic data in various sizes. Besides, we evaluated the method on a wider column of data to evaluate the scalability. In addition, we measured and made a comparison of the execution on three different architectures: Vector Machine, an up-to-date GPGPU, and x86 CPU, along with various compilers and BLAS implementations.