Adaptive multiclass support vector machine for multimodal data analysis. (October 2017)

• Record Type: Journal Article
• Authors: Zhang, Xin; Zhang, Xiu
• Is Part Of: Pattern recognition. Volume 70(2017:Oct.)
• Page Start: 177
• Page End: 184
• Publication Date: 2017-10
• Subjects: Artificial bee colony -- Feature selection -- Hyperparameter optimization -- Multiclass classification -- Support vector machine; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.05.006
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Multimodal data commonly exists in human lives. Early analysis usually concentrates on mining information based on single modality. Recent studies show that learning tasks could be greatly enhanced by analyzing data from the aspect of multimodality. This paper deals with classifying multimodal data comprised of visual and acoustic contents. Different data features are fused under a hierarchical structure to achieve a good semantic understanding. Then, to accomplish accurate classification, an adaptive support vector machine method (ASVM) is proposed. The method is support vector machine with hyperparameters controlled by a novel and efficient artificial bee colony algorithm. First, a micro colony is set as the number of hyperparameters is usually less than 5. Second, one position inheritance based on roulette wheel selection is used. Third, discarded solutions are mutated by position shift operation instead of random reinitialization. The ASVM method is first verified on classical data sets demonstrating the goodness of the proposed method. Then the proposed method is applied on a multimodal data set. Each sample includes both image and audio data features. Experimental results show that the ASVM method is more effective and robust than the compared methods.