Support vector machines (SVMs) have proved to be highly successful for use in many applications that require classification of data. However, training an SVM requires solving an optimization problem that is quadratic in the number of training examples. This is increasingly becoming a bottleneck for SVMs because while the size of the datasets is increasing, especially in applications such as bioinformatics, single-node processing power has leveled off in recent years. One possible solution to these trends lies in solving SVMs on multiple computing cores or on computing clusters.
In my talk I will show two approaches to solving SVMs in parallel, one based on holding the complete kernel matrix in distributed memory and the other on training an ensemble of smaller SVMs in parallel. I will compare these solvers to a popular single-node solver. The comparison covers accuracy, speed, and the ability to process large datasets. I will show that while none of these solvers performs well on all three metrics, each of them ranks high on two of them.
Finally, I will describe IBM’s Parallel Machine Learning toolbox which allows practitioners to rapidly implement parallel learning algorithms.
*Project URL:* _www.alphaworks.ibm.com/tech/pml_
*Speaker home page:* _http://domino.research.ibm.com/comm/research_people.nsf/pages/yomtov.ind…