BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Contributed Talk: Discovery of Spin-Crossover Metal-Organic Framew orks from Limited and Noisy Data using Quantile Active Learning - Virtual Presentation - Roberta Poloni (Université Grenoble Alpes) DTSTART:20260212T100000Z DTEND:20260212T103000Z UID:TALK242341@talks.cam.ac.uk DESCRIPTION:Data-driven materials discovery is often hindered when target properties are computationally expensive or experimentally demanding to ob tain\, making conventional large-scale screening impractical. This challen ge is particularly acute for metalâ\;&euro\;&ldquo\;organic framework s (MOFs)\, whose vast chemical diversity and complex electronic behavior d emand both accuracy and data efficiency. Here\, we present a unified activ e learning strategy based on regression tree methods to accelerate the dis covery of functional MOFs under scarce\, noisy\, and imbalanced data condi tions.\nUsing low-dimensional\, physically motivated descriptors derived f rom stoichiometric and geometric features\, we construct regression tree&a circ\;&euro\;&ldquo\;based partitions of the descriptor space to actively select the most diverse and informative samples for electronic-structure e valuation. This new approach\, that we name Regression Treeâ\;&euro\; &ldquo\;Active Learning [1]\, is demonstrated across multiple MOF datasets \, where it yields compact training sets that outperform existing active l earning strategies in predicting band gaps\, adsorption properties\, and o ther key materials descriptors\, while exhibiting reduced variance and enh anced robustness to uneven label distributions [2].\nWe further apply this framework to the discovery of spin-crossover (SCO) MOFs\, a rare but tech nologically promising subclass relevant for sensing\, spintronics\, and ga s-related applications. By coupling a new Quantile Regression Treeâ\; &euro\;&ldquo\;Active Learning approach with Random Forest regression and new density functional theory calculations\, necessary to predict this pro perty\, we accurately identify SCO-active candidates from limited and impe rfect training data\, recovering over 80% of true positives. This strategy enables the identification of a new set of high-confidence SCO MOFs\, dem onstrating that complex quantum phenomena can be reliably uncovered throug h data-efficient\, actively guided exploration of large materials spaces [ 2].\n \;\n[1] Data Min Knowl Disc (2023)\;\n[2] J. Am. Chem. Soc. 2024 \, 146\, 9\, 6134â\;&euro\;&ldquo\;6144\, https://doi.org/10.1021/jac s.3c13687\;\n[2] npj Comput. Mater.\, submitted. LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR