optimization and uncertainty quantification at exascale - Michael McKerns

optimization and uncertainty quantification at exascale

Michael McKerns ／英語

    We have developed and implemented a comprehensive, rigorous and algorithmic framework capable of utilizing all available information to rigorously predict the impact of high-impact rare events, where our predictors are multiply-nested global optimizations over all possible valid scenarios. Such optimizations are high-dimensional, highly-constrained, non-convex, and generally impossible to solve with current optimization technology; however, by addressing optimization constraints as quantum operators on a probability distribution, our software converts highly-nonlinear statistical calculations to those that are nearly embarrasingly parallel.  By utilizing abstractions on programming models and global distributed caching of data and results, we can scale up from desktop calculations to petascale and larger with little burden on the programmer, where the upscaling depends on the complexity of the information at hand (constraints, restraints, theoretical models, datasets, etc).
    

關於講者

Mike has over fifteen years of teaching experience in physics, applied math, and computing, and has taught twenty financial and science workshops in the past year alone.  He been a research scientist at Caltech since 2002, where he has served as manager and lead developer for two $15M software projects on predictive science and large-scale computing. In the past five years, his software has been the backbone of several research projects on large-scale risk analysis and predictive science.  Mike is a co-founder of the UQ Foundation, a non-profit for the advancement of predictive science, and co-creator of OUQ theory, a rigorous mathematical framework for uncertainty quantification. Mike has a B.S. in Applied Physics from Notre Dame, and a Ph.D. in Physics from the University of Alabama Birmingham. He has been developing parallel and distributed computing software infrastructure for ten years, and large-scale optimization and risk analysis software frameworks for over five years.  His software frameworks have been used within HPC credit risk applications, at national labs on spallation neutron sources, and are being used as validation and performance testing suites in the design of the first generation of exescale computers. His software has over 15,000 total downloads to unique IP addresses, including one library available in Fedora and RHEL distributions.