【摘要】 分析了多实例学习(MIL)在复杂数据目标(图像,基因)等方面的广泛应用,针对大多数已存在的MIL算法仅能处理小样本或中等规模样本的问题,为了处理MIL中的大规模问题,提出了一种高效可扩展的MIL集成学习算法——B2VMI(Bag to Vector Multi-instance)。该集成学习算法利用低计算成本的映射方法,将传统的MIL包映射成新的特征向量表示,以此方式获得包级信息。在多个多实例数据集上的实验表明,B2VMI具有可扩展等优秀性能,该算法不仅能够取得同当前先进的MIL集成学习算法可比较的精确度,而且具有比其他MIL集成学习算法快5倍的效率。更多还原

【Abstract】 Multi-instance Learning( MIL) has been widely used in a variety of applications,such as complex data objects( image,genes,etc.). However,most existing MIL algorithms can only handle small samples or medium-sized samples. To deal with the large-scale MIL problem,it provided an efficient and extensible MIL ensemble learning algorithm B2 VMI( Bag to vector multi-instance) in this paper. The ensemble learning algorithm uses the low-cost calculation method to map the traditional MIL bag into a new eigenvector representation to obtain the bag level information in this way. Experiments on multiple multi-instance datasets showed that B2 VMI has excellent performance such as scalability. This algorithm can not only achieve the same accuracy as the current advanced MIL ensemble learning algorithms,but also the efficiency of the algorithm is faster than other MIL ensemble learning algorithms 5 times.更多还原