Title: Machine Learning Tools for Uncertain Grids Method Simulations

Speaker: Sergey A. Terekhov, Ph.D., Chief Science Officer, Neurok Software, LLC

Date/Time: Monday, February 23, 2009 at 2:00 – 3:00 pm

Location: CSRI Building/Room 90 (Sandia NM)

Brief Abstract: This short presentation is intended to stimulate a discussion concerning the rapidly evolving area of Machine Learning and its potential value in the context of simulations using the novel Uncertain Grids Method as well as other mathematical modeling approaches.

The presentation will begin with a brief overview of machine learning research and civil applications at the Artificial Neural Networks Laboratory of the All-Russian Scientific Research Institute of Technical Physics (VNIITF). The main part of the presentation will focus on particular applications and current areas of research in machine learning which have the potential to significantly enhance the analysis of simulation results and contribute to the overall success of computational modeling.  Topics will include:

• Machine learning (e.g. neural) approximations and probabilistic models for coefficients (EOS, boundary conditions, sources etc.)
• Non-linear response surface models for series of explorative simulations (both certain and probabilistic)
• Machine learning models as "surrogates" for target function and constraints during design optimization
• Stochastic field models for quantifying uncertainties in physics and/or modeled system geometry

Dr. Terekhov is a talented mathematician and physicist specializing in the application of neural networks, genetic algorithms and other soft computing technologies to various aspects of data mining, machine learning and data-driven inference.

For 16 years Dr. Terekhov worked at the All-Russian Scientific Research Institute of Technical Physics (VNIITF), eventually becoming the founder and manager of VNIITF’s Artificial Neural Networks Laboratory. He has authored or co-authored more than 50 papers on topics such as plasma physics, artificial neural networks and applications.

Since joining Neurok Software in 2001, Dr. Terekhov has managed several machine learning projects in the field of neural pattern recognition for non-destructive evaluation of oil pipelines, intelligent control system development and analysis of experimental data in oil engineering. He is also a featured lecturer at the annual meetings of the Russian Neural Networks Society

CSRI POC: Randall M. Summers, (505) 844-6296