CUDA-based acceleration of KNN and Bayesian algorithms

Abstract:
Machine learning algorithms are data-driven algorithms that have produced significant advancements in areas like natural language processing, image classification, and machine translation. The computation time, on the other hand, climbs exponentially as the dataset grows. This paper proposes a method of utilizing CUDA to employ GPU parallel computing to cut the computational complexity of frequently used algorithms such as the K-nearest neighbor (KNN) and Naive Bayes (NB), in an effort to expand the algorithm's computational performance. On the basis of assuring the algorithm's correctness, this method considerably reduces the algorithm's running time. Experiments show that the method proposed in this paper can reduce the complexity of KNN from O(N2) to O(N), and also the complexity of the NB algorithm from O(N) to O(1).