Labeling Complicated Objects: Multi-View Multi-Instance Multi-Label Learning∗

Multi-Instance Multi-Label (MIML) is a learning framework where an example is associated with mul-tiple labels and represented by a set of feature vec-tors (multiple instances). In the formalization of MIML learning, instances come from a single source (sin-gle view). To leverage multiple information sources (multi-view), we develop a multi-view MIML frame-work based on hierarchical Bayesian Network, and de-rive an effective learning algorithm based on variational inference. The model can naturally deal with exam-ples in which some views could be absent (partial ex-amples). On multi-view datasets, it is shown that our method is better than other multi-view and single-view approaches particularly in the presence of partial exam-ples. On single-view benchmarks, extensive evaluation shows that our method is highly competitive or better than other MIML approaches on labeling examples and instances. Moreover, our method can effectively handle datasets with a large number of labels