I started at Google six months ago tomorrow, working on the team writing Google's infrastructure for federated learning.
Federated learning is a technology enabling machine learning applications without the need to centralize data. It addresses a problem space that is expanding seemingly without limit right now, and is an excellent example of breaking down the false barriers demarcating tradeoffs that we are told are fundamental.
Federated learning gets its name from the computational model it abstracts--federations of devices collaborating on a machine learning model. Federated computations are more general than this instance, however--you can imagine federted analytics, computing statistics over some distributed dataset without grabbing the data itself.
I have been working at Google on a software framework for describing these federated computations, which we open-sourced in February as TensorFlow Federated (TFF). TFF is under active development, taking up most-to-all of my working time.
I've settled a little better into our Seattle life, including a new dog named Erwin (for the great physicist), and I hope to be able to maintain this site a little better. Take this as a quick update on what I have been up to, and a plan to take a little more care with this site in the future.