Multiple Instance Learning

Más información sobre el libro

This book offers a comprehensive overview of multiple instance learning (MIL), outlining its framework and key paradigms. The authors delve into essential MIL algorithms, including classification, regression, and clustering, with a particular emphasis on classification. A taxonomy is established, highlighting significant proposals and developing efficient algorithms to extract relevant information amidst uncertainty. Key applications are presented, along with an exploration of related fields such as distance metrics and alternative hypotheses. Chapters address emerging aspects of MIL, including data reduction for multi-instance issues and the challenge of imbalanced MIL data, defined at the bag level. This representation leverages ambiguity since bag labels are known, but individual instance labels remain undefined. The book also investigates multiple instance multiple label learning, which introduces flexibility and ambiguity in object representation, allowing a single object to be represented by a bag of instances with multiple associated class labels. This resource is ideal for developers and engineers applying MIL techniques to tackle real-world challenges, and it serves as a valuable reference for researchers and students seeking an in-depth understanding of MIL literature, methods, and tools.