IMCN seminar - Dr. Gareth John CONDUIT

Royal Society University Research Fellow
University of Cambridge

Abstract :

We present a deep learning tool that merges all possible sources of information: computer simulations, physical laws, hidden variables, and even noise.  The tool can exploit all possible sources of information to make the best possible predictions.  We illustrate the tool through case studies that include experimental validation and have all led to commercialization: we design a nickel-base alloy for direct laser deposition from a training set containing just ten core results by juxtaposing complementary material properties; a molybdenum-base alloy for a forging hammer; a smallmolecule bioactivity database was initially 0.05% complete is imputed to be 30% complete, which is then used to design new drugs; and we present further examples spanning healthcare to infrastructure.