Multi-dimensional Classification via Selective Feature Augmentation

Abstract In multi-dimensional classification (MDC), the semantics of objects are characterized by multiple class spaces from different dimensions. Most MDC approaches try to explicitly model dependencies among in output space. contrast, recently proposed feature augmentation strategy, which aims at manipulating space, has also been shown be an effective solution for MDC. However, existing only focus on designing holistic augmented features appended with original features, while better generalization performance could achieved exploiting kinds features. this paper, we propose selective strategy that focuses synergizing Specifically, assuming part is pertinent and useful each dimension’s induction, derive a can fully utilize conduct selection To validate effectiveness generate three simple based standard k NN, weighted maximum margin techniques, respectively. Comparative studies show achieves superior against both state-of-the-art its degenerated versions either kind