One of the first real-world experiments carried out with SVM was done at AT&T Bell Labs using optical character recognition (OCR) data ([17,54]). These early experiments already showed the astonishingly high accuracies for SVMs which was on a par with convolutive multi-layer perceptrons. Below we record the classification performance of SVM, some variants not discussed in this chapter, and other classifiers on the USPS (US-Postal Service) benchmark (parts from ). The task is to classify handwritten digits into one of ten classes. Standard SVMs as presented here already exhibit a performance similar to other methods.
A benchmark problem larger than the USPS data set ( patterns) was collected by NIST and contains handwritten digits. Invariant SVMs achieve an error rate of () on this challenging and more realistic data set, slightly better than the tangent distance method () or single neural networks (LeNet : ). An ensemble of LeNet networks that was trained on a vast number of artificially generated patterns (using invariance transformations) is on a par with the best SVM and also achieved ().
The high dimensional problem of text categorization seems to be an application for which SVMs have performed particularly well. A popular benchmark is the Reuters- text corpus, where Reuters collected news stories from 1997, and partitioned and indexed them into different categories to simplify the access. The feature typically used to classify Reuters documents are -dimensional vectors containing word frequencies within a document. With such a coding SVMs have achieved excellent results (see e.g. ).
A challenging problem of computational chemistry is the discovery of active chemical compounds. The goal is to found the compounds that bind to a certain molecule in as few iterations of biological testing as possible. An astonishingly effective application of SVM to this problem has been recently proposed by .
There are numerous other applications to which SVM were successfully applied. Examples are object and face recognition tasks (), inverse problems ([71,77]), gene array expression monitoring (), remote protein homology detection (), or splice site detection ([79,64]). This list can be extended.