Reduced sets and fast approximation for kernel methods

Kernel methods often need much computational time in their testing stages due to large numbers of support vectors, especially in dealing with some large datasets. This paper presents two reduced set approaches - reduced set selection (RSS) and reduced set construction (RSC) to fast approximate the kernel methods, and compares their performances on the benchmark repository. Experimental results demonstrate that both the two approaches can speed up the kernel methods greatly, while RSC behaves better than RSS in the most cases.