Indexing Multiple-Instance Objects

As an actively investigated topic in machine learning, Multiple-Instance Learning (MIL) has many proposed solutions, including supervised and unsupervised methods. We introduce an indexing technique supporting efficient queries on Multiple-Instance (MI) objects. Our technique has a dynamic structure that supports efficient insertions and deletions and is based on an effective similarity measure for MI objects. Some MIL approaches have proposed their similarity measures for MI objects, but they either do not use all information or are time consuming. In this paper, we use two joint Gaussian based measures for MIL, Joint Gaussian Similarity (JGS) and Joint Gaussian Distance (JGD). They are based on intuitive definitions and take all the information into account while being robust to noise. For JGS, we propose the Instance based Index for querying MI objects. For JGD, metric trees can be directly used as the index because of its metric properties. Extensive experimental evaluations on various synthetic and real-world data sets demonstrate the effectiveness and efficiency of the similarity measures and the performance of the corresponding index structures.