Conditional Restricted Boltzmann Machines for Multi-label Learning with Incomplete Labels
Xin Li, Feipeng Zhao, Yuhong Guo · 2015
Standard multi-label learning methods as-sume fully labeled training data. This as-sumption however is impractical in many ap-plication domains where labels are difficult to collect and missing labels are prevalent. In this paper, we develop a novel condi-tional restricted Boltzmann machine model to address multi-label learning with incom-plete labels. It uses a restricted Boltzmann machine to capture the high-order label de-pendence relationships in the output space, aiming to enhance the capacity of recover-ing missing labels and learning high quality multi-label prediction models. Moreover, it also incorporates label co-occurrence infor-mation retrieved from auxiliary resources as prior knowledge. We perform model training by maximizing the regularized marginal con-ditional likelihood of the label vectors given the input features, and develop a Viterbi style EM algorithm to solve the induced optimiza-tion problem. The proposed approach is eval-uated on four real word multi-label data sets by comparing to a number of state-of-the-art methods. The experimental results show it outperforms all the other comparison meth-ods across the applied data sets. 1