As part of the module “Advanced topics in machine learning”, which I am attending as part of my MPhil at Cambridge, I had to give a presentation on the paper “Grammar Variational Autoencoder“(GVAE). The paper presents an autoencoder which improves the state-of-the-art for discrete structures, by using the idea that very often discrete structures can be generated from a syntax free grammar. They give the examples of molecules and arithmetic expressions. If you are interested in learning more, check out my presentation.