SVM-Cuda

We use simplified SMO algorithm to achieve SVM training process

Naive Version

Folder simple-smo contains the sequential version of the code.
Use g++ to compile and run the code.

CUDA Version

Folder cuda-svm contains the cuda version of the code.
Run compile_all.sh to compile the code.
Run cuda_svm_v0 to see speedup by pre-compute the Kernel Matrix using CUDA based on naive version.
Run cuda_svm_v1 to see speedup by Calculate the Error Matrix using CUDA based on v0.
Run cuda_svm_latest to see speedup by Calculate only the updates of Error Matrix using CUDA based on v1.

Authors

• Zekun Fan
• Hanzhi Chen
• Site Li