SciANN: A TensorFlow API for scientific computations with neural networks

April 28, 2021 | 1:00-2:00pm CT (Webinar) | 2:00-3:00pm CT (Workshop)

About the Webinar

Over the past decade, artificial neural networks, also known as deep learning, have revolutionized many computational tasks, including image classification and computer vision, search engines and recommender systems, speech recognition, autonomous driving, and healthcare. Even more recently, this data-driven framework has made inroads in engineering and scientific applications, such as earthquake detection, fluid mechanics and turbulence modeling, dynamical systems, and constitutive modeling.

A recent class of deep learning known as physics-informed neural networks (PINN), where the network is trained simultaneously on both data and the governing differential equations, is particularly well suited for solution and inversion of equations governing physical systems, in domains such as fluid mechanics, solid mechanics and dynamical systems. This increased interest in engineering and science is due to the increased availability of data and open-source platforms such as Theano, TensorFlow, and PyTorch, which offer features such as high-performance computing and automatic differentiation.

In the first part of this presentation, I describe the application of deep learning, as a unified framework, for scientific computations including curve fitting and regression, solving ordinary and partial differential equations, and model inversion and calibration. I will then introduce you to SciANN, a TensorFlow API designed to perform such operations efficiently. In the second part of the presentation, we will perform a live hands-on tutorial of SciANN using the DesignSafe system.

About the Presenter

Dr. Ehsan Haghighat is a Mitacs Postdoctoral Fellow at UBC studying stochastic modeling and uncertainty quantification of engineering systems, and a Simulation Consultant for Seismix Reservoir Management LLC. Previously, he was a Postdoctoral Associate at MIT where he studied the assessment of induced seismicity due to CO2 sequestration and oil and gas injection and production, Stochastic Modeling, and Machine Learning. He received his Ph.D. from McMaster University specializing in Computational Geomechanics. After his Ph.D., he worked at Forming Technologies Inc (currently Hexagon AB), as a mechanics developer and project leader, where he developed implicit FEM solver for modeling the sheet metal forming process. His research interests include computational methods for mechanics of solids and porous media, stochastic modeling and uncertainty quantification, and machine learning of engineering systems.

Webinar Registration