Speaker: David S. Matteson
Professor and Associate Department Chair of Statistics and Data Science,
Cornell University
Website: https://davidsmatteson.com/
Wednesday, April 24, 2024
11:30 am: Lunch (Kitchen)
12:00 pm: Talk (Seminar Room #1327)
at the Yale Institute for Foundations of Data Science, Kline Tower, 13th Floor
Webcast Link (via Zoom – starts at 12:00): https://yale.zoom.us/s/7859884026
Title: Drift vs Shift: Decoupling Trends and Changepoint Analysis
Abstract: We introduce a new approach for decoupling trends (drift) and changepoints (shifts) in time series. Our locally adaptive model-based approach for robustly decoupling combines Bayesian trend filtering and machine learning based regularization. An over-parameterized Bayesian dynamic linear model (DLM) is first applied to characterize drift. Then a weighted penalized likelihood estimator is paired with the estimated DLM posterior distribution to identify shifts. We show how Bayesian DLMs specified with so-called shrinkage priors can provide smooth estimates of underlying trends in the presence of complex noise components. However, their inability to shrink exactly to zero inhibits direct changepoint detection. In contrast, penalized likelihood methods are highly effective in locating changepoints. However, they require data with simple patterns in both signal and noise. The proposed decoupling approach combines the strengths of both, i.e. the flexibility of Bayesian DLMs with the hard thresholding property of penalized likelihood estimators, to provide changepoint analysis in complex, modern settings. The proposed framework is outlier robust and can identify a variety of changes, including in mean and slope. It is also easily extended for analysis of parameter shifts in time-varying parameter models like dynamic regressions. We illustrate the flexibility and contrast the performance and robustness of our approach with several alternative methods across a wide range of simulations and application examples.
Bio: David S. Matteson is Professor of Statistics & Data Science at Cornell University and Director of the National Institute of Statistical Sciences (NISS.org) – the joint research unit of the American Statistical Association (ASA), the International Biometric Society (IBS), the Institute of Mathematical Statistics (IMS). At Cornell, he is also Associate Department Chair of Statistics & Data Science, Director of Graduate Studies in Data Science, and Founding Executive Committee Member for the Cornell Center for Data Science for Enterprise & Society. His research centers on methods and theory for modeling complex human and natural processes and systems, with expertise in dynamic, spatial, functional, and network data science and machine learning. He received a CAREER Award from the National Science Foundation (NSF), the Chancellor’s Award for Scholarship and Creative Activities from the State University of New York, the inaugural Ann S. Bowers Research Excellence Award, and Faculty Research Awards from the Xerox/PARC Foundation and LinkedIn. He has served as lead PI and Director for several NSF funded collaborative institutes, and he founded and serves as Editor-in-Chief for the new open access ASA affiliated journal Data Science in Science.
