BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Steven Torrisi (Harvard University) DTSTART:20210723T160000Z DTEND:20210723T170000Z DTSTAMP:20260423T005816Z UID:CRIBB/9 DESCRIPTION:Title: Wh ich parts matter? Interpretable random forest models for X-Ray absorption spectra\nby Steven Torrisi (Harvard University) as part of Computatio nal Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nX-ray abs orption spectroscopy (XAS) produces a wealth of information about the loca l structure of materials\, but interpretation of spectra often relies on e asily accessible trends and prior assumptions about the structure. Recentl y\, researchers have demonstrated that machine learning models can automat e this process to model the environments of absorbing atoms from their XAS spectra. However\, machine learning models are often difficult to interpr et\, making it challenging to determine when they are valid and whether th ey are consistent with physical theories. In this work\, we present three main advances to the data-driven analysis of XAS spectra: we demonstrate t he efficacy of random forests in solving two new property determination ta sks (predicting Bader charge and mean nearest neighbor distance)\, we addr ess how choices in data representation affect model interpretability and a ccuracy\, and we show that multiscale featurization can elucidate the regi ons and trends in spectra that encode various local properties. The multis cale featurization transforms the spectrum into a vector of polynomial-fit features\, and is contrasted with the commonly-used “pointwise” featu rization that directly uses the entire spectrum as input. We find that acr oss thousands of transition metal oxide spectra\, the relative importance of features describing the curvature of the spectrum can be localized to i ndividual energy ranges\, and we can separate the importance of constant\, linear\, quadratic\, and cubic trends\, as well as the white line energy. \n\nThis work has the potential to assist rigorous theoretical interpreta tions\, expedite experimental data collection\, and automate analysis of X AS spectra\, thus accelerating the discovery of new functional materials. We expect that this featurization strategy could be useful for broad domai ns of application\, such as one-dimensional time-series analysis or other forms of spectroscopy.\n\nPaper: https://www.nature.com/articles/s41524-02 0-00376-6\n\n============================================================= ==============\n\nZOOM MEETING info:\n\n https://mit.zoom.us/j /96155042770\n\n Meeting ID: 961 5504 2770\n LOCATION:https://researchseminars.org/talk/CRIBB/9/ END:VEVENT END:VCALENDAR