|About this Abstract
||Materials Science & Technology 2019
||Data Science for Material Property Interpretation
||A Machine Learning Approach to Independent Component Analysis for Nuclear Magnetic Resonance Spectra
||Ryan J. McCarty, Jason Ernest Kahn, David Rotter, Ryan Padilla, Nadia Ahmed, Domingos R. Begalli
|On-Site Speaker (Planned)
||Ryan J. McCarty
Nuclear magnetic resonance spectroscopy (NMR) is a valuable technique for understanding the atomic structures of solids, but applying NMR to materials containing mixtures of chemical compounds results in complex spectra that are difficult to interpret. Independent component analysis (ICA) techniques allow algorithm-based decompositions of input spectra into statistically relevant components. Ideally, these components should resemble the individual spectra of real chemical compounds, but the quality of the produced components depends on the intuition and technical expertise of the scientist responsible for selecting inputs and preferred algorithms. This process can result in iterative selection until outputs match a preconceived idea of “good” results.
To standardize the accuracy of predicted components and variability of individual judgment, we benchmark models constructed using machine learning techniques trained on synthetic NMR spectra to traditional ICA methods to find the best fit components of multi-phase mixtures.
||Definite: At-meeting proceedings