Understanding research papers can be hard. Paper authors typically assume that you share the same context and prior knowledge that they have. When you don’t, it becomes hard to follow the paper. Looking for material from which to understand all of the references you don’t understand can be tedious, and the quality of different materials varies greatly. Sometimes it’s not even clear what to start searching to clarify parts of the paper.
We at DFL share an interest in developing a deeper understanding. Together, we chose a few modern papers that we believed were important in inspiring new ideas and research. We spent some time understanding each paper and writing down the core concepts on which they were is built. We then found the best material to learn each concept, added exercises to practice our understanding, and met weekly to discuss each of the previous week’s readings so that we could rigorously understand the content.
We did this across two locations, one at Google AI in Mountain View as part of the Google AI Residency and the other at New York University as part of Joan Bruna’s class on The Mathematics of Deep Learning. We realized that the resources we had compiled would be useful to a much broader community, and this repository of depth-first study plans is the outcome of that genesis.
DFL was designed to be shared with and grown by the community. It is not limited to people at Google or at NYU. If you are interested in developing your own curriculum for a paper that we do not already host, we will help you! Check out out github for examples and open an issue detailing which paper you want to study. We want your help in growing this resource and making it richer.