Minisymposium Presentation

We present our work on unveiling microscopic details of colloidal and soft matter systems through a novel integration of small-angle neutron scattering (SANS), molecular simulations, computations, and machine learning (ML). First, we demonstrate how ML was employed to invert the scattering of charged colloidal particles to their relevant structural parameters. Molecular dynamics simulations, a probabilistic Gaussian process framework, and a variational autoencoder were trained, and a trained decoder was iteratively applied to fit the input scattering experiment data, thereby closing the loop by transforming experimental SANS data into structural parameters. Similarly, we applied a methodology involving Monte Carlo simulations of AB-type diblock copolymers with excluded volume effects at the dilute limit, utilizing an ML framework of the Gaussian process to inversely determine the conformation of these copolymers from their coherent scattering. Finally, we introduce our newly developed deep learning inversion framework that employs convolutional neural networks to accurately extract morphological features from a model lamella-forming system based on its SANS spectra.