Rapid developments in communications, networking, AI robots, 3D printing, genomics, blockchain, novel materials, and powerful computation platforms are rapidly bringing data-generating people, processes and devices together. The interactions between data analytics in multiple regimes (sparse, panel, big data, etc.) and other fields are exciting because the tools that are being invented now may enable new, faster and semi-automated methods of scientific discovery. These, in turn, might further amplify the pace of progress and significantly impact various areas of health, science and technology.
Motivated by the above, my research has been focused on the modeling, representation, detection and prediction from data. In this seminar, I will introduce some of my past research under the framework of a universal data analytic engine "TimeHunter" that I have been developing. In particular, I will introduce some new foundational principles and efficient algorithms in the areas of model selection, nonlinear time series, change detection, and multi-regime analysis. I will highlight the key idea of each contribution, and demonstrate their performance with both synthetic and real data.