Knowledge Amalgamation for Multi-Label Classification via Label Dependency Transfer

Multi-label classification (MLC), which assigns multiple labels to each instance, is crucial domains from computer vision text mining. Conventional methods for MLC require huge amounts of labeled data capture complex dependencies between labels. However, such datasets are expensive, or even impossible, acquire. Worse yet, these pre-trained models can only be used the particular label set covered in training data. Despite this severe limitation, few exist expanding predicted by models. Instead, we acquire vast new and retrain a model scratch. Here, propose combining knowledge (teachers) train student that covers union teachers. This supports broader than any one its teachers without using We call problem amalgamation multi-label classification. Our method, Adaptive KNowledge Transfer (ANT), trains learning teacher’s partial infer global all across show ANT succeeds unifying among teachers, outperforming five state-of-the-art on eight real-world datasets.