An empirical study on multi-instance learning

作者：Chen Sisi; Jiang Liangxiao

来源：Advances in Information Sciences and Service Sciences, 2012, 4(6): 193-202.

DOI：10.4156/AISS.vol4.issue6.23

• Classification algorithm
• Empirical studies
• Multi-instance learning
• Objective functions
• Three categories
• Traditional learning

Different from traditional learning frameworks: supervised learning, unsupervised learning and reinforcement learning, multi-instance learning is a new framework of learning and it's a variation of supervised learning, where the task is to learn a concept given positive and negative bags of instances. This paper focus on the classification algorithms of multi-instance learning and these algorithms can be divided into three main categories based on different ways of studying: 1. Algorithms that studying specially for multi-instance learning. 2. Algorithms that importing the restrain of multi-instance learning to single-instance learning algorithms' objective functions. 3. Algorithms that transforming the multi-instance learning problem to the single-instance problem. Based on these three categories, we empirically do three groups experiments to observe their performance and then draw conclusions. The experimental results verify our conclusions drawn in this paper.

• 出版日期2012