AM Seminar: Data-driven modeling of stochastic systems using physics-aware deep learning

Speaker Name: 
Paris Perdikaris
Speaker Title: 
Assistant Professor
Speaker Organization: 
University of Pennsylvania
Start Time: 
Monday, April 15, 2019 - 4:00pm
End Time: 
Monday, April 15, 2019 - 5:00pm
Location: 
BE 372
Organizer: 
Daniele Venturi
Abstract:

We present a probabilistic deep learning methodology that enables the construction of predictive data-driven surrogates for stochastic systems. Leveraging recent advances in variational inference, we put forth a scalable computational framework for discovering surrogate models from paired input-output observations of a system that may be stochastic in nature, originate from different information sources of variable fidelity, or be corrupted by complex noise processes. We also show how physical constraints can be employed as informative priors that introduce a regularization mechanism for effectively constructing robust deep learning models in cases where the cost of data acquisition is high and training data-sets are typically small. The effectiveness of the proposed methods is demonstrated through a series of canonical studies involving stochastic dynamical systems and nonlinear conservation laws. 

Short Bio: 
Paris' works spans a wide range of areas in computational science and engineering, with a particular focus on the analysis and design of complex physical and biological systems using machine learning, stochastic modeling, and high-performance computing. Prior to Penn, he spent two years as a post-doctoral researcher at MIT, developing machine learning algorithms that synergistically combine multi-fidelity data with prior knowledge (e.g., differential equations), towards establishing a new paradigm in predictive modeling and decision making under uncertainty. 
 
Education: 

Ph.D. - Applied Mathematics, Brown University (2015)