Organizers: Ehsan Elhamifar
Amit Roy-Chowdhury
Amin Karbasi
Description: The increasing amounts of data in computer vision requires robust and scalable summarization tools to efficiently extract most important information from massive datasets. However, summarization involves optimization programs that are nonconvex and NP-hard, in general. While convex, nonconvex and submodular optimization have been studied intensively in mathematics, successful and effective applications of them for information summarization along with new theoretical results have recently emerged. These results, in contrast with more classical approaches, can deal with struc-tured data, nonlinear models, data nuisances and exponentially large dataset. The goal of this tutorial is to present the audi-ence with a unifying perspective of this problem, introducing the basic concepts and connecting nonconvex methods with convex sparse optimization as well as submodular optimiza-tion. The presentation of the formulations, algorithms and theoretical foundations will be complemented with applica-tions in computer vision, including video and image summari-zation, procedure learning from instructional data, pose esti-mation, active learning and more.
Schedule:
1345 Overview of Summarization Algorithms: Modeling, Optimizations, Applications, Ehsan Elhamifar
1430 Submodular Optimization Methods for Summarization, Amin Karbasi
1530 Afternoon Break
1600 Sequential Data Summarization and Applications to Procedure Learning, Ehsan Elhamifar
1645 Collaborative Summarization With Side Information, Amit Roy-Chowdhury
