报告题目：Redesigning Support Vector Machines on Modern Hardware
报 告 人：文泽忆博士（新加坡国立大学）
A survey conducted by Kaggle in 2017 shows that 26% of the data mining and machine learning practitioners use Support Vector Machines (SVMs) to solve their problems. However, SVM training and prediction are very expensive for large and complex problems. In this talk, I will present our recent project called ThunderSVM which exploits the high-performance of Graphics Processing Units (GPUs) and multi-core CPUs. ThunderSVM provides efficient algorithms for classification (SVC), regression (SVR) and one-class SVMs. It supports all the functionalities of LibSVM and uses identical command line options as LibSVM, such that existing LibSVM users can easily apply ThunderSVM. Multiple language interfaces including Python, R and Matlab are also supported. Our experimental results show that ThunderSVM is generally an order of magnitude faster than LibSVM while producing identical SVMs. Moreover, ThunderSVM can run on machines without GPUs. Documentations and source code are available on the project website at https://github.com/zeyiwen/thundersvm.
Dr. Zeyi Wen is a Research Fellow in National University of Singapore. Before working in Singapore, Zeyi was a research fellow at University of Melbourne from 2015 to 2016, completed his PhD degree at University of Melbourne in 2015, and received a bachelor's degree in Software Engineering from South China University of Technology in 2010. His areas of research include Data Mining, Machine Learning and High-performance Computing.