A deep-learning approach to molecular dynamics
Deep learning—the combination of computer-controlled laser pulses and modern ab initio calculations running on GPU computers—will greatly accelerate our ability to probe and control chemistry. Just like chemical reagents are used for testing, chemically transforming, and sensing other compounds, photons could serve the same purpose. The key challenge is determining how to create ‘photonic reagents’ that have the desired selectivity. The Dantus Research Group has been developing computer-controlled laser pulses to probe and control laser-matter interactions. This talk will illustrate the power of deep learning as applied to chemical reactions of organic compounds proceeding through complex mechanisms involving a roaming H2 molecule. Results will also be presented on how computer-controlled pulses are being developed for improving medical diagnosis through spectroscopic biomedical imaging.