基于弱标签的多示例迁移学习方法

作为监督学习的一种变体,多示例学习(MIL)试图从包中的示例中学习分类器。在多示例学习中,标签与包相关联,而不是与单个示例相关联。包的标签是已知的,示例的标签是未知的。MIL可以解决标记模糊问题,但要解决带有弱标签的问题并不容易。对于弱标签问题,包和示例的标签都是未知的,但它们是潜在的变量。现在有多个标签和示例,可以通过对不同标签进行加权来近似估计包和示例的标签。提出了一种新的基于迁移学习的多示例学习框架来解决弱标签的问题。首先构造了一个基于多示例方法的迁移学习模型,该模型可以将知识从源任务迁移到目标任务中,从而将弱标签问题转换为多示例学习问题。在此基础上,提出了一种求解多示例迁移学习模型的迭代框架。实验结果表明,该方法优于现有多示例学习方法。

Multi-instance transfer learning method based on weak labels

As a variant of supervised learning,MIL attempts to learn classifiers from instances in bags.In multi-instance lear-ning,labels are associated with bags,not individual instances.The label of bag is known,the label of instance is unknown.MIL can solve the label ambiguity problem,but it is not easy to solve the problem of weak labels.For the weak labels problem,the labels of bags and instances are unknown,but they are potential variables.Now that there are multiple labels and instances,it can approximate estimate the labels of bags and instances by weighting the different weak labels.This paper proposed a new transfer learning-based multi-instance learning(TMIL)framework based on transfer learning to solve the problem of weak labels.Firstly,it constructed a transfer learning model based on the multi-instance method,which could transfer knowledge from the source task to the target task,thus transforming the weak labels problem into a multi-instance learning problem.On this basis,it proposed an iterative framework to solve the multi-instance transfer learning model.Experimental results show that this method is superior to the existing multi-instance learning method.