Towards Super-resolution Molecular AI-Art, 2020

    The capability of machine learning in the form of generative adversarial networks (GANs) to autonomously construct complex patterns and fill-in gaps in imagery opens the potential for visually connecting molecular and cellular scales through the creation of very large, science-based artworks. 

    Towards this goal, a GAN was repurposed from Github tools (Inkawhich S.) and trained on datasets created from the works of 1) scientist/artist Dr. David Goodsell (right sets), and 2) supercomputer simulations from the RIKEN institute (below).

    This is a work in progress, and subsequent steps towards the ultimate goal of creating ‘wall-sized, molecular resolution human cells’ with AI will involve the integration of additional methods, such as pluralistic image completion (Zheng C. et al., 2019) and progressive growth (Beers A. et al., 2018).