Random forests for metric learning with implicit pairwise position dependence

• DOI: 10.1145/2339530.2339680
• Publication Date: 2012-08-14T08:18:56Z
• Authors: Caiming Xiong; David Johnson; Ran Xu; Jason J. Corso
• Keywords: Mahalanobis distance; Feature vector; Feature (linguistics)

Metric learning makes it plausible to learn semantically meaningful distances for complex distributions of data using label or pairwise constraint information. However, date, most metric methods are based on a single Mahalanobis metric, which cannot handle heterogeneous well. Those that multiple metrics throughout the feature space have demonstrated superior accuracy, but at severe cost computational efficiency. Here, we adopt new angle problem and is able implicitly adapt its distance function space. This adaptation accomplished by random forest-based classifier underpin incorporate both absolute position standard relative into representation. We implemented tested our method against state art global multi-metric variety sets. Overall, proposed outperforms types in terms accuracy (consistently ranked first) an order magnitude faster than (16x worst case).