In this project, we team up with many students from other laboratories to attend KDD cup 2010. We were divided into small teams, competed with other teams and gave presentations every week. By doing so, every team can absorb other teams experience and polish their model. Finally, by using ensemble methods, we won the first prize of all teams.
This is our GPGPU(General-purpose computing on graphics processing units) project in Graduate School. Its purpose is to dig the possibility of performance improvement of machine learning algorithms by the power of CUDA. In 2010, CUDA is a relative new technique. Through three simple machine learning algorithms: linear kernel SVM, K-means, Neural Network, we find out that it is possible to improve machine learning algorithms' efficiency by paralleling some component of machine learning algorithms.