The Interplay Between Big Data and Sparsity in Systems Identification: Some Lessons from Machine Learning*. Issue 28 (2015)

• Record Type: Journal Article
• Authors: Cheng, Y.; Wang, Y.; Camps, O.; Sznaier, M.
• Is Part Of: IFAC-PapersOnLine. Volume 48:Issue 28(2015)
• Page Start: 1285
• Page End: 1292
• Publication Date: 2015
• Subjects: Dynamic Big Data -- Sparse Optimization -- Identification of Switched Systems -- Regularized Regression; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2015.12.309
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: The past few years have witnessed an unprecedented growth in data collection capabilities. When coupled with recent advances in distributed control, these developments have the potential to make feasible a large range of applications, such as self-aware, smart environments, that can profoundly impact society. However, realizing this potential, requires developing identification techniques that can handle very large data sets, whose dimensionality challenges the capabilities of existing techniques. The goal of this paper is to present a tutorial illustrating how this \curse of dimensionality" can be combated by exploiting the inherent sparsity exhibited by a large class of identification problems. While many of the ideas linking big data and sparsity appeared originally in the context of machine learning, porting these ideas to dynamic settings presents both new challenges and opportunities. The paper concludes by briey exploring these opportunities, and in particular the contributions that the systems identification community can make to the problem of inferencing in data deluged scenarios.