Evolving kernels for support vector machine classification

While support vector machines (SVMs) have shown great promise in supervised classification problems, researchers have had to rely on expert domain knowledge when choosing the SVM's kernel function. This project seeks to replace this expert with a genetic programming (GP) system. Using strongly typed genetic programming and principled kernel closure properties, we introduce a new algorithm, called KGP, which finds near-optimal kernels. The algorithm shows wide applicability, but the combined computational overhead of GP and SVMs remains a major unresolved issue. Categories and Subject Descriptors I.5.1 [Computing Methodologies]: Pattern Recognition--Statistical General Terms Experimentation Keywords Genetic Programming, Support Vector Machines

• Type: Conference
• Year: 2007
• Where: GECCO
• Authors: Keith Sullivan, Sean Luke
• Keywords: GECCO 2007, Genetic Programming, Optimization, Support Vector Machines, SVM's Kernel Function