Optimizing resources in model selection for support vector machines

Tuning SVM kernel parameters is a an important step for achieving a high-performing learning machine. The usual automatic methods used to tune these parameters require an inversion of the Gram-Schmidt matrix or a resolution of an extra quadratic programming problem. In the case of a large dataset these methods require the addition of huge amounts of memory and a long CPU time to the already significant resources used in the SVM training. In this paper, we propose a fast method based on an approximation of the gradient of the empirical error along with incremental learning, which reduces the resources required both in terms of processing time and of storage space.