February 24, 2020: Three Short Stories: Tales of Data-Driven Science
- February 24, 2020
- 4:00 p.m.
- 100 Hancock Hall
- Dr. Erik Bollt, Clarkson University
- Faculty Host: Dr. Jonathan Black
Abstract: An important growing niche of the data science-machine learning revolution combines the best of the new massive data processing powers thanks to modern algorithms and hardware, with the informed perspective of the human expert’s prior knowledge underlying scientific principles. These learning technologies do not replace the skills of the scientist and engineer, but they can empower us to advance new capabilities to thrive in today’s massively over-sensed, data rich environment. We offer here a sample of machine learning informed physics and discovery for AOE to demonstrate capabilities and practices. Our first story contrasts how low-dimensionality that may well describe certain complex spatiotemporal dynamics allows reduced order modelling from common linear modal analysis to nonlinear manifold learning methods, specialized to the intrinsic geometry and physical details of the problem. Our second short story regards the contrasting problems of transport versus coherence and patterns by adapting spectral methods for transfer operators applied here to interpret oceanographic systems. Finally our last story highlights concepts from inverse problems for remote sensing in Earth sciences when the scientist already has a good general idea of the kind dynamics that drives the system. Building such knowledge directly into the machine aided discovery process not only improves results, but validates and informs the belief in the underlying processes.
Bio: Erik Bollt, (PhD, University of Colorado, Boulder) is a Full Professor at Clarkson University and endowed as the W. Jon Harrington Professor of Mathematics with appointment to Electrical and Computer Engineering, ECE, and serves as Director of C3S2 the Clarkson Center for Complex Systems Science. Professor Bollt specializes in data enabled science with machine learning and dynamical systems. Professor Bollt has recently published a book on these topics as applied to systems such as the Gulf of Mexico oil spill, (Applied and Computational Measurable Dynamics, Book Publisher: Society for Industrial and Applied Mathematics, (2013)).