Efficient Implementation of Coreset-based K-Means Methods

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2021-06-14

Department

Major/Subject

Xiaobo Wu

Mcode

SCI3044

Degree programme

Master’s Programme in Computer, Communication and Information Sciences

Language

en

Pages

53

Series

Abstract

The problem of efficiently extracting information from big data has received significant attention in both academia and industry. In the last decades, many computational models like distributed/parallel computing, streaming algorithms and dynamic algorithms have been proposed to cope with big data. These methods were commonly used nowadays and have obtained great performances in many scenarios related to dealing with a massive amount of data. However, they are algorithm-driven approaches which means you need to generate new ideas every time, and it still will be constrained by the growing data size. Recently, a data-driven approach named coreset was developed to generate a much smaller summary to the original big data and directly apply the existing algorithms in the small dataset to reach an acceptable result in a very decent time and space complexity. In the latest research, the size of coreset $|S|$ can be handle to uncorrelated with the original data size $|D|$, which means unlimited by the size of data. My main result is an efficient implementation of an open-source coreset-based modern computational framework based on C++ and CUDA programming language in this thesis project. Specifically, we implement a coreset library that can construct a small coreset for arbitrary shapes of numerical data with a decent time cost. My job was mainly to follow the research proposed by Braverman et al. (SODA 2021). The experiments were conducted to benchmark with the CuML library, a well-known collection of the efficient GPU implementation of various machine learning algorithms. The benchmark results prove our implementation of the coreset method can lead to a faster speed in solving the k-means problem among the big data with high accuracy in the meantime. In addition, by utilizing the composable property of the coreset, our merge-and-reduce version of the coreset breaks through the restriction of GPU memory.

Description

Supervisor

Jiang, Shaofeng

Thesis advisor

Jiang, Shaofeng

Keywords

coreset, k-means, cuda, cuml, c++, machine learning

Other note

Citation