Affiliation:
1. Science College , Heilongjiang Bayi Agricultural University , Daqing , Heilongjiang , , China
Abstract
Abstract
Partial least squares are the most widely used identification algorithm, but the algorithm cannot achieve real-time performance for massive data. To solve this application contradiction, a parallel computing strategy based on NVIDIA CU-DA architecture is proposed to implement the partial least squares algorithm using a graphics processor (GPU) with massively parallel computing features as the computing device and combining the advantages of GPU memory. Research and analysis found that the partial least squares algorithm implemented using CUDA on GPU is 48 times faster than the implementation of the CPU. Therefore, the algorithm has good usability and higher application value, which makes it possible to apply the partial least squares algorithm to massive data statistics.
Subject
Applied Mathematics,Engineering (miscellaneous),Modeling and Simulation,General Computer Science
Reference22 articles.
1. Ren, H., Zou, C., & Li, R. (2022). Extrapolation-based Tuning Parameters Selection in Massive Data Analysis. SCIENTIA SINICA Mathematica, 52(6), 689-.
2. Murtagh, F. (2017). Massive Data Clustering in Moderate Dimensions from the Dual Spaces of Observation and Attribute Data Clouds.
3. Zhu, R. (2015). Poisson Subsampling Algorithms for Large Sample Linear Regression in Massive Data. Stats.
4. Pan, R., Zhu, Y., Guo, B., et al. (2021). A Sequential Addressing Subsampling Method for Massive Data Analysis under Memory Constraint. arXiv e-prints.
5. Zhao, Y. (2018). Feasible Algorithm for Linear Mixed Model for Massive Data. Communications in Statistics, B. Simulation and Computation.