I've been working on a deep learning project that feels relevant to this community, and would love to hear some of your thoughts. I'm training a neural network to translate abstract syntax trees into images (and then back into ASTs since the whole system is an autoencoder.) Right now this is limited to simple arithmetic statements like

c / (a+b)

, but the goal is to try it out on larger expressions with a greater vocabulary, so that you can turn larger ASTs into images. In addition to being really fun and interesting, it feels like it could offer ideas for visual programming? Visualized here is an interpolation between different expression trees. Still kind of rough around the edges.