Research And Lmplementation Of Parallelization Of Machine Learning Algorithm Based On CUDA Platform

At present,the main task of machine learning is to study the analysis of high-volume user data to help managers make decisions.Due to the current data dimension and the large number of samples,CPU serial processing is time consuming.On the other hand,GPU(Graphics Processing Unit)rapid development,with a strong parallel processing capabilities.As the GPU is cheap and cheap and natural parallel,researchers began to use the GPU to do the general calculation.CUDA(Compute Unified Device Architecture)is NVIDIA's launch for the NVIDIA GPU to achieve the general computing power of the programming environment.Using the CUDA programming model,we can use the GPU to parallelize the design and implementation of the algorithm for machine learning.This thesis mainly studies the feasibility and implementation of GPU parallelization of basic machine learning algorithm,hoping to find a general transplant scheme from CPU platform to CUDA platform.The main tasks include:Aiming at the classification machine learning algorithm,taking KNN and decision tree algorithm as an example,the performance consumption module of the original algorithm is analyzed firstly,then the CUDA acceleration of the main performance consumption module is carried out.Finally,the KNN and decision tree algorithm are designed for the parallelization scheme of CUDA,and the KNN algorithm is selected The experiment,comparative analysis of parallel differences before and after.Finally,the clustering machine learning algorithm based on CUDA parallelization is summarized.Aiming at the algorithm of clustering machine learning,taking k-means and DBScan as an example,the performance consumption module of the original algorithm is analyzed firstly,then the CUDA acceleration of the main performance consumption module is carried out.Finally,the k-means and DBScan algorithms are designed for CUDA parallelization scheme,Means algorithm,and the difference between before and after parallelization is analyzed.Finally,the clustering machine learning algorithm based on CUDA parallelization is summarized.At the end of this thesis,the CUDA-based machine learning parallelization scheme is successfully applied to practical engineering.