Data Analytics and Physics - Informed Models: The Next Generation Takes Shape
Monday, 20 July 2020, 3:25 p.m.–5:10 p.m. | Austin, Texas
3:25 Introductory Remarks
3:40 Hecto Klie, Chief Executive Officer and Lead Scientist, DeepCast
3:50 Cedric Fraces Gasmi, Director, Tenokonda Inc and Stanford University
4:00 Ruben Rodriguez Torrado, Founder and Chief Executive Officer, OriGen.ai
4:10 Rami M Younis, Associate Professor, McDougall School of Petroleum Engineering, Director, Future Reservoir Simulation Systems & Technology (FURSST) University of Tulsa
4:20 Moderated Panel
4:45 Audience Q&A
In recent years a substantial debate has developed between predictive machine learning models versus theoretically-correct physics models. Out of this debate has arisen a phenomena known as a physics-informed predictive model. A number of researchers in this field are claiming that this type of model seems to have adopted the best of both worlds. This panel will describe this kind of model, its benefits, typical pitfalls, and give examples of the latest case studies.
Domain experts are more important than ever as machine learning algorithms are being pushed to the limit, particularly for quantifying model uncertainty and risk assessment. Decisions can be made on the basis of a predictive model, but those decisions can be deeply flawed if data accuracy and physics representation aspects were not considered. We will discuss the key considerations in today’s decision-making environment, particularly in a situation where there is likely to be large-scale changes in ownership, and the new operators of fields will be under pressure to develop data-driven models that are both reliable and quickly modified. Tactics for data managing and feature engineering, as well as identifying key pitfalls in developing the algorithms, and deciding which algorithms to employ will be discussed.
Fee: Included with Registration