(186s) Graph Versus Images : Nanowire Network Prediction

Networks are fundamental representations of various complex systems. Self-assembled nanowire networks, in particular, exhibit diverse patterns that correspond to different physical properties. Traditional observation and analysis methods, such as scanning electron microscopy (SEM) and optical microscopy, are both time and resource-intensive. These techniques often fail to capture the complete structure of the networks. Therefore, accurate prediction of nanowire networks is crucial for quality control and property prediction in nanofilm production. In this study, we utilize SEM images of localized areas of self-assembled silver nanowire networks to predict structures in unexplored regions. Our prediction approach involves three generative model based on three types of data input: network features, network graphs, and network images. While feature and image-based predictions using machine learning are widespread, our research introduces a novel integration of graph synthesis methods and generative models to predict network graphs. This technique effectively addresses the challenge of predicting graphs of varying sizes. Our work significantly advances the field of subnetwork graph prediction, with substantial implications for nanomaterials research and beyond.