Optimization of SVM Kernels and Application to Down Category Recognition

摘要

In recent years, the use of support vector machines (SVMs)on various classifications has been increasingly popular.However, the results of classification usually depend on the parameters of the model. These parameters are usually picked by experience, experimental compare, and large-scale search or cross-validation provided by software package. Our scheme to optimize SVM hyper-parameters is to minimize an empirical error estimate using a Quasi-Newton optimization method on the validation set. The method has been used successfully in our down category recognition system.