Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
Kernel ensemble support vector machine with integrated loss in shared parameters space
Ist Teil von
Multimedia tools and applications, 2023-05, Vol.82 (12), p.18077-18096
Ort / Verlag
New York: Springer US
Erscheinungsjahr
2023
Link zum Volltext
Quelle
SpringerLink (Online service)
Beschreibungen/Notizen
In this paper, we propose a kernel ensemble SVM with integrated loss in shared parameters space. Different from the traditional multiple kernel learning methods of seeking linear combinations of basis kernels as a unified kernel, the proposed method aims to find multiple solutions in corresponding Reproducing Kernel Hilbert Spaces (RKHSs) simultaneously. To achieve this goal, we draw on the idea of multi-view data processing, and the individual kernel gram matrix is considered as one view of the data. We, therefore, propose an ensemble idea to combine those multiple individual kernel losses into a whole one through an integrated loss design. Therefore, each model can co-optimize to learn its optimal parameters by minimizing the integrated loss in multiple RKHSs. Besides, another feature of our method is the introduction of shared and specific parameters in multiple RKHSs for learning. In this manner, the proposed model can learn the common and individual structures of the data from its parameters space, thereby improving the accuracy of the classification task and further enhancing the robustness of the proposed ensemble model. Experimental results on several UCI classification and image datasets demonstrate that our method performs best among state-of-the-art MKL methods, such as SimpleMKL, EasyMKL, MREKLM, and MRMKL.