Multi-instance multi-label learning in the presence of novel class instances

Multi-instance multi-label learning (MIML) is a framework for learning in the presence of label ambiguity. In MIML, experts provide labels for groups of instances (bags), instead of directly providing