21 May 2021, 16:00 Athens time via Zoom
Variable-memory models for inference and learning with discrete signals
Discrete time series, often consisting of streaming "big" data sets, are becoming very common in modern biological, medical, communications, and financial applications. Consequently, there is increasingly pressing demand for methods that can perform inferential and learning tasks in a real-time, sequential fashion. We will describe how the family of context tree sources, a class of variable-memory models studied extensively in information theory, can be used for an extremely broad array of statistical and learning tasks across a variety of applications. We will show how the context tree weighting (CTW) and related algorithms, developed by Willems, Shtarkov, Tjalkens and their collaborators since the early 1990s, can be generalized and extended to provide methodological and algorithmic tools for effective Bayesian inference on discrete time series. After describing the new, general Bayesian setting, we will present both deterministic and simulation-based procedures for several tasks, including prediction, model selection, change-point detection, and estimation. Results illustrating the performance of the resulting methods on real and synthetic data sets will be presented.
About the speaker
Ioannis Kontoyiannis received the B.Sc. degree in mathematics from Imperial College, University of London, in 1992, a distinction in Part III of the pure mathematics tripos from Cambridge University in 1993, and the M.S. degree in statistics and the Ph.D. degree in electrical engineering from Stanford University in 1997 and 1998, respectively. From 1998 to 2001, he was with Purdue University. From 2000 to 2005 he was with the Division of Applied Mathematics, Brown University, where he was also with the Department of Computer Science. From 2005 to 2021, he was with the Department of Informatics, Athens University of Economics and Business. From 2018 to 2020, he was a Professor with the Department of Engineering, University of Cambridge, where he held the Chair of Information and Communications and was the Head of the Signal Processing and Communications Laboratory. Since 2020, he has been the Churchill Professor of mathematics with the Department of Pure Mathematics and Mathematical Statistics, University of Cambridge. Dr. Kontoyiannis was awarded the Manning endowed Assistant Professorship in 2002 and was awarded an Honorary Master of Arts degree Ad Eundem, in 2005, both by Brown University. In 2004, he was awarded a Sloan Foundation Research Fellowship. He is a Fellow of the IEEE.