CVPR 2018 Tutorial on Big Data Summarization

CVPR 2018 tutorial on Big Data Summarization: Algorithms and Applications.

The increasing amounts of data in computer vision requires robust tools to extract most important information from large collections of data. The summarization problem addresses this challenge by finding a small subset of most informative data points from large datasets. However, summarization often leads to optimization programs that are nonconvex and NP-hard. While (non)convex programming and submodular optimization have been studied intensively in mathematics, successful and effective applications of them for the problem of information summarization along with new theoretical results have recently emerged. These results, in contrast with more classical approaches, can deal with structured data, nonlinear models, data nuisances and exponentially large dataset. The goal of this tutorial is to present the audience with a unifying perspective of this problem, introducing the basic concepts and connecting nonconvex methods with convex sparse optimization and submodular optimization. The presentation of the formulations, algorithms and theoretical foundations will be complemented with applications in computer vision, including video and image summarization, procedure learning from instructional data, pose estimation, active learning and more.