PROJECT TITLE :

Heterogeneous Metric Learning of Categorical Data with Hierarchical Couplings - 2018

ABSTRACT:

Learning applicable metric is crucial for effectively capturing complex knowledge characteristics. The metric learning of categorical data with hierarchical coupling relationships and local heterogeneous distributions is very difficult nonetheless rarely explored. This Project proposes a Heterogeneous mEtric Learning with hIerarchical Couplings (HELIC for brief) for this kind of categorical knowledge. HELIC captures each low-level worth-to-attribute and high-level attribute-to-class hierarchical couplings, and divulges the intrinsic heterogeneities embedded in every level of couplings. Theoretical analyses of the effectiveness and generalization error bound verify that HELIC effectively represents the above complexities. Intensive experiments on 30 data sets with numerous characteristics demonstrate that HELIC-enabled classification considerably enhances the accuracy (up to forty.ninety three %), compared with 5 state-of-the-art baselines.