A Review of Latest Multi-instance Learning

Due to the application needs of some special scenarios, multi-instance learning problem has been paid more and more attention in recent years. Different from the traditional supervised learning problem, each example in the training set of multi-instance learning is not represented by a single feature vector, but a group of feature vectors, the example is called bag, and the vector contained in them are called instances. Multi-instance learning is widely used in many real scenarios. Therefore, it has become an important topic in machine learning, and many algorithms related to multi-instance learning have been proposed successively. In this paper, the latest applications of multi-instance learning in some real scenarios are described in detail, the main ideas of some new multi-instance learning algorithms are given, and finally summarizes and prospects.