Multiple Instance Learning
Multiple Instance Learning (MIL) is a type of weakly supervised learning, where we attempt to learn a concept from a training set of labeled bags, where each bag contains multiple unlabeled instances. As result of such unusual task formulation MIL has a special property – it allows to detect local patterns knowing only class labels assigned globally. This also can be used for results and model interpretation. In this report I will review several MIL algorithms and how they can be applied in practice. As it turned out MIL fits well for diverse application fields such as computer vision and text processing, especially well it was used in medical image and video analysis tasks.