作者: Yuan Jiang , De-Chuan Zhan , Han-Jia Ye , Yang Yang
DOI:
关键词:
摘要: In real world applications, data are often with multiple modalities. Previous works assumed that each modality contains sufficient information for target and can be treated equal importance. However, it is different modalities of various importance in tasks, e.g., the facial feature weak fingerprint strong ID recognition. this paper, we point out should strategies propose Auxiliary Regularized Machine (ARM), which by extracting most discriminative subspace while regularizing modal predictor. Experiments on binary multi-class datasets demonstrate advantages our proposed approach ARM.