Vicki Boykis put together a free primer (delivered in PDF) on vector space embeddings.
Peter Norvig urges us to teach ourselves programming in ten years. In this spirit, after several years of working with embeddings, foundational data structures in deep learning models, I realized it’s not trivial to have a good conceptual model of them. Moreover, when I did want to learn more, there was no good, general text I could refer to as a starting point. Everything was either too deep and academic or too shallow and content from vendors in the space selling their solution. So I started a project to understand the fundamental building blocks of machine learning and natural language processing, particularly as they relate to recommendation systems today. The results of this project are the PDF on this site, which is aimed at a generalist audience and not trying to sell you anything except the idea that vectors are cool. I’ve also been working on Viberary to implement these ideas in practice.
The post also points to plenty of follow on educational material. I’m particularly intrigued by the supplemental content related to deep learning and recommender systems.