首页|University of Manchester Reports Findings in Machine Learning (An Unsupervised M achine Learning Approach for the Automatic Construction of Local Chemical Descri ptors)
University of Manchester Reports Findings in Machine Learning (An Unsupervised M achine Learning Approach for the Automatic Construction of Local Chemical Descri ptors)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-New research on Machine Learning is th e subject of a report. According to news reporting originating in Manchester, Un ited Kingdom, by NewsRx journalists, research stated, "Condensing the many physi cal variables defining a chemical system into a fixed-size array poses a signifi cant challenge in the development of chemical Machine Learning (ML). Atom Center ed Symmetry Functions (ACSFs) offer an intuitive featurization approach by means of a tedious and labor-intensive selection of tunable parameters." The news reporters obtained a quote from the research from the University of Man chester, "In this work, we implement an unsupervised ML strategy relying on a Ga ussian Mixture Model (GMM) to automatically optimize the ACSF parameters. GMMs e ffortlessly decompose the vastness of the chemical and conformational spaces int o well-defined radial and angular clusters, which are then used to build tailormade ACSFs. The unsupervised exploration of the space has demonstrated general a pplicability across a diverse range of systems, spanning from various unimolecul ar landscapes to heterogeneous databases. The impact of the sampling technique a nd temperature on space exploration is also addressed, highlighting the particul arly advantageous role of high-temperature Molecular Dynamics (MD) simulations. The reliability of the resulting features is assessed through the estimation of the atomic charges of a prototypical capped amino acid and a heterogeneous colle ction of CHON molecules. The automatically constructed ACSFs serve as high-quali ty descriptors, consistently yielding typical prediction errors below 0.010 elec trons bound for the reported atomic charges. Altering the spatial distribution o f the functions with respect to the cluster highlights the critical role of symm etry rupture in achieving significantly improved features. More specifically, us ing two separate functions to describe the lower and upper tails of the cluster results in the best performing models with errors as low as 0.006 electrons. Fin ally, the effectiveness of finely tuned features was checked across different ar chitectures, unveiling the superior performance of Gaussian Process (GP) models over Feed Forward Neural Networks (FFNNs), particularly in low-data regimes, wit h nearly a 2-fold increase in prediction quality. Altogether, this approach pave s the way toward an easier construction of local chemical descriptors, while pro viding valuable insights into how radial and angular spaces should be mapped."
NETL
NSTL
