Analogy-preserving Semantic Embedding for Visual Object Categorization
Sung Ju Hwang, Kristen Grauman, Fei Sha · 2013
In multi-class categorization tasks, knowl-edge about the classes ’ semantic relationships can provide valuable information beyond the class labels themselves. However, existing techniques focus on preserving the seman-tic distances between classes (e.g., according to a given object taxonomy for visual recog-nition), limiting the influence to pairwise structures. We propose to model analogies that reflect the relationships between mul-tiple pairs of classes simultaneously, in the form “p is to q, as r is to s”. We translate se-mantic analogies into higher-order geometric constraints called analogical parallelograms, and use them in a novel convex regularizer for a discriminatively learned label embed-ding. Furthermore, we show how to dis-cover analogies from attribute-based class de-scriptions, and how to prioritize those likely to reduce inter-class confusion. Evaluating our Analogy-preserving Semantic Embedding (ASE) on two visual recognition datasets, we demonstrate clear improvements over exist-ing approaches, both in terms of recognition accuracy and analogy completion. 1.