BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:Europe/Stockholm X-LIC-LOCATION:Europe/Stockholm BEGIN:DAYLIGHT TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20220812T074334Z LOCATION:Nairobi Room DTSTART;TZID=Europe/Stockholm:20220628T120000 DTEND;TZID=Europe/Stockholm:20220628T123000 UID:submissions.pasc-conference.org_PASC22_sess143_msa125@linklings.com SUMMARY:Representation and Optimization of Materials with Deep Learning DESCRIPTION:Minisymposium\n\nRepresentation and Optimization of Materials with Deep Learning\n\nGomez-Bombarelli\n\nGiven adequate training data, ma chine learning models trained over experimental or theoretical outcomes en able virtual screening and inverse design of molecules and materials. Howe ver, activity data is typically expensive and slow to acquire. Thus, findi ng representations of atomistic structure that allow learning structure-ac tivity relationships that are fast, robust and data-efficient is key.
Here, we will show our recent work in representation learning and design of materials. In particular, we will report 3D and 4D deep learning model s that allow conformation information to achieve better transferability. F urthermore, we well describe hierarchical macromolecule representations fo r biomacromolecules and graph neural network architectures fine-tuned for property prediction tasks in crystalline materials. Multi-fidelity models mixing in computational and experimental data improve predictive accuracy without added computational cost. Once robust predictive models are traine d, a variety of tools become available for design. We will describe the ap plication of genetic-algorithm and Monte Carlo Tree Search strategies to f ind therapeutic macromolecules and effective solid catalysts respectively, in combination with\n\nDomain: Chemistry and Materials, Physics END:VEVENT END:VCALENDAR