Predicting Soft Matter Evolution Using Machine Learning

Soft matter including colloids, polymers and granular material display behavior and self-organization that are difficult and sometimes impossible to predict due to complex interactions with the environment, which itself can change and self-organize. The traditional approach to understand these problems is to study structural evolution by theory and experiment. We are taking the different approach of machine learning. We build convolutional neural networks (CNNs) to process the large amounts of data. We are applying these new tools to direct imaging in rheo-optics, to fitness evaluation of cell growth when they pass near obstacles, and to ecological microsystems with soft matter flavor. The common element is to predict properties varying with time.