MIL(multi-instance learning)

MIL算法在71 mil基准数据集上进行了测试。这是最大的实验MIL存储库，用于算法比较。数据集的应用领域是分子活性预测，图像注释，文本分类，网页分类和音频记录分类 (n miproblems.org提供数据集的mat文件)。
每个数据集文件都是一个逗号分隔值 (CSV) 格式的文件，它具有实例数许多行和特征数许多列以及两个附加列。第一个附加列对应于传播到实例的bag类标签。第二列是bag ID列，其中每个实例接收其所有者bag的bag id号。其余列分别存储实例的特征值。

The MIL algorithm was tested on 71 MIL benchmark datasets, which form the largest experimental MIL repository for algorithm comparison. The application domains of these datasets include molecular activity prediction, image annotation, text classification, web classification, and audio recording classification. The MAT-formatted datasets are available at miproblems.org. Each dataset file is in comma-separated values (CSV) format, with rows equal to the number of instances, columns equal to the number of features, and two additional columns. The first additional column stores the class label of the bag to which the instance belongs. The second column is the bag ID column, where each instance is assigned the bag ID of its parent bag. The remaining columns contain the feature values of each instance.

搜集汇总

背景与挑战

背景概述

MIL数据集是一个用于多示例学习算法测试的基准数据集集合，包含71个CSV格式的数据文件，应用于分子活性预测、图像注释等多个领域。数据集发布于2018年，由multipleinstancelearning.com提供。