Professor Todd Martinez, Stanford University. Leveraging Machine Learning and Stream Processors for Theoretical Chemistry. R.B. Woodward Lectures in the Chemical Sciences, Greater Boston Theoretical Chemistry Seminar.
Abstract: Novel computational architectures and methodologies are revolutionizing diverse areas ranging from video gaming to advertising and espionage. In this talk, I will discuss how these tools and ideas can be exploited in the context of theoretical and computational chemistry. I will discuss how insights gleaned from recommendation systems (such as those used by Netflix and Amazon) can lead to reduced scaling methods for electronic structure, how the algorithms in electronic structure can be adapted for commodity stream processing architectures such as graphical processing units, and how nonlinear dimensionality reduction methods can be used to extract chemical knowledge from the resulting data. I will discuss some of the details regarding the structure and implementation of these methods. I will also show how these advances can be harnessed to progress from traditional “hypothesis-driven” methods for using electronic structure and first principles molecular dynamics to a “discovery-driven” mode where the computer is tasked with discovering chemical reaction networks. Finally, I will show how these can be combined with force-feedback (haptic) input devices and three-dimensional visualization to create molecular model kits that carry complete information about the underlying electrons. This interactive first principles molecular dynamics method (molecular computer-aided design or mol-CAD) opens the door to novel ways of teaching chemistry and may also be of use in applied chemical research.
LOCATION:MIT, Room 4-163 STATUS:CONFIRMED DTSTART:20140409T200000Z DTEND:20140409T213000Z END:VEVENT END:VCALENDAR