Strata Data 2017 Recap
Last week, we got the opportunity to present Set a couple of public events, Strata being by far the biggest. This was my third or fourth time presenting at a Strata event and they continue to impress me with their quality.
This year, our presentation was titled, “Learning location: Real-time feature extraction for mobile analytics”. The primary focus was our on-line data sketching techniques for learning and querying long-term behavioral data directly on a mobile device without the need for data transfer to the server. We shared a few new open-source libraries we use for building data sketches in Swift.
We also were able to talk a little bit about how this approach allows us to do machine learning on the device and gives us the ability to do fully anonymous machine learning off of the device. This is really exciting for us, since it allows us to leverage collectively trained neural nets that are distributed and evaluated on the device. Finally, we touched on our over-the-air app configurations through what we call, Recipes. This is one of the most exciting areas of our work and we will write it up more completely in the near future.