MIC-SVM: Designing a Highly Efficient Support Vector Machine for Advanced Modern Multi-core and Many-Core Architectures

Yang You, Shuaiwen Leon Song, Haohuan Fu, Andres Marquez, Maryam Mehri Dehnavi, Kevin Barker, Kirk W. Cameron, Amanda Peters Randles, Guangwen Yang
2014 2014 IEEE 28th International Parallel and Distributed Processing Symposium  
Support Vector Machine (SVM) are widely used in datamining and big data applications. In recent years, SVM has been used in High Performance Computing (HPC) for power/performance prediction, auto-tuning, and runtime scheduling. However, to avoid runtime training overhead HPC researchers can only afford to apply offline model training. This often leads to loosing prediction accuracy because no runtime information is available. Advanced multi-and many-core architectures offer massive parallelism
more » ... ith complex memory hierarchies which makes runtime training possible, but efficiently parallelizing the SVM algorithm on these architectures is challenging. To address the challenges above, we have designed and implemented MIC-SVM, a highly efficient parallel SVM for x86 based multi-core and many-core architectures, such as the Intel Ivy Bridge CPUs and Intel Xeon Phi co-processor (MIC). We propose various novel analysis methods and optimization techniques to fully utilize the multilevel parallelism provided by these architectures. The proposed techniques serve as general optimization methods for other machine learning tools. MIC-SVM achieves 4.4-84x and 18-47x speedup against the popular LIBSVM, on MIC and Ivy Bridge CPUs respectively, for several realworld data-mining datasets. Compared to GPUSVM, ran on a modern NVIDIA k20x GPU, the performance of our MIC-SVM is competitive. We also conduct a cross-platform performance comparison, focusing on Ivy Bridge CPUs, MIC, and GPUs. We also provide insights on how to select the most suitable architecture for individual algorithms and input data patterns.
doi:10.1109/ipdps.2014.88 dblp:conf/ipps/YouSFMDBCRY14 fatcat:vv6jzxoccbek5hz7bgiqhyt3dq