{"id":185,"date":"2017-02-13T18:04:12","date_gmt":"2017-02-13T18:04:12","guid":{"rendered":"http:\/\/sag.art.uniroma2.it\/kelp_wordpress\/?page_id=185"},"modified":"2017-04-05T15:38:10","modified_gmt":"2017-04-05T15:38:10","slug":"kernel-functions-2","status":"publish","type":"page","link":"http:\/\/www.kelp-ml.org\/?page_id=185","title":{"rendered":"Kernel functions"},"content":{"rendered":"

Kernel methods (see for example (Shawe-Taylor and Cristianini, 2004)) are a powerful class of algorithms for pattern analysis that, exploiting the so called kernel functions, can operate in an implicit high-dimensional feature space without explicitly computing the coordinates of the data in that space. Most of the existing machine learning platforms provide kernel methods that operate only on vectorial data. On the contrary, KeLP\u00a0has the fundamental advantage that there is no restriction on a specific data structure, and kernels directly operating on vectors, sequences, trees, graphs, or other structured data can be defined.<\/p>\n

Furthermore, another appealing characteristic of KeLP\u00a0is that complex kernels can be created by composing and combining simpler kernels. This allows to create richer similarity metrics in which different information from different Representation<\/a>s can be simultaneously exploited.<\/p>\n

As shown in the figure below, KeLP completely supports the composition and combination mechanisms providing three abstractions of the Kernel<\/a> class:<\/p>\n