Kernel matrix plays an important role in kernel methods. The kernel matrix contains all of the topology information included in the training sample set, which inspired us to select the significant samples out of a large data set by the kernel matrix. Giving a certain kernel function, the learning efficiency almost depends on the size of the training sample set. This means that once we shrink the scale of training samples, the learning process can be remarkably sped up. In the paper, a sampling algorithm based on the alignment is proposed to improve the calculation efficiency, where kernel alignment is an approach to measure the similarity between different kernel functions and matrices. More specifically, our algorithm can be utilized to pick out the significant samples from any large data set in unsupervised learning problems. The learning efficiency can be greatly improved adopting the small-scaled selected sample set. Another advantage of our method is that the novel sampling process can be completed online, because the algorithm is ready for the update of the training set. We applied the algorithm to detect the defects on tickets' surface. The experimental results indicate that our sampling algorithm not only reduces the mistake rate but also shortens the detection time.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.