A Multiple-Instance Learning Based Approach to Multimodal Data Mining

A Multiple-Instance Learning Based Approach to Multimodal Data Mining

Zhongfei (Mark) Zhang (SUNY Binghamton, USA), Zhen Guo (SUNY Binghamton, USA) and Jia-Yu Pan (Google Inc., USA)
DOI: 10.4018/978-1-4666-0900-6.ch007


This paper presents multiple-instance learning based approach to multimodal data mining in a multimedia database. This approach is a highly scalable and adaptable framework that the authors call co-learning. Theoretic analysis and empirical evaluations demonstrate the advantage of the strong scalability and adaptability. Although this framework is general for multimodal data mining in any specific domain, to evaluate this framework, the authors apply it to the Berkeley Drosophila ISH embryo image database for the evaluations of the mining performance in comparison with a state-of-the-art multimodal data mining method to showcase the promise of the co-learning framework.
Chapter Preview

In the machine learning community, MIL has become a focused topic in recent years and has received extensive attention in the literature ever since the classic work of (Dietterich et al., 1997; Auer, 1997; Maron &Lozano-Perez, 1998). Recent developments on MIL include (Andrews et al., 2003; Andrews & Hofmann; 2004; Rahmani & Goldman, 2006). (Yang & Lozano-Perez, 2000) and (Zhang et al., 2002) were among the first to apply MIL to image retrieval, which led to more subsequent work on this topic (Zhang et al., 2006; Zhu et al., 2006).

(Chen et al., 2006) recently added the embeded instance selection principle into the classic MIL algorithm resulting in a better learning performance, and also applied this method to image retrieval.

Complete Chapter List

Search this Book: