Supervised and Unsupervised Machine Learning of Structural Phases of Polymers Adsorbed to Nanowires

We present our work and findings on identifying configurational phases and structural transitions in a polymer-nanotube composite using a variety of machine learning methods. We employ dimensionality reduction, conventional neural networks, and the confusion method, a more recent neural-network-based approach. We find neural networks are able to reliably recognize all polymer structures that have previously been found in experiment and simulation. Furthermore, we are able to locate boundaries between configurational phases in a way that does not rely on preconceived, ad-hoc order parameters.