Thanks to David Westbrook, Peter Brandt, Samuel Ieong, Chenyu Zhao, Li Wei, Michalis Potamias, Evan Rosen, Barry Rosenberg, Christine Robson, James Pine, Tal Shaked, Tushar Chandra, Mustafa Ispir, Jeremiah Harmsen, Konstantinos Katsiapis, Glen Anderson, Dan Duckworth, Shishir Birmiwal, Gal Elidan, Su Lin Wu, Jaihui Liu, Fernando Pereira, and Hrishikesh Aradhye for many corrections, suggestions, and helpful examples for this document. Also, thanks to Kristen Lefevre, Suddha Basu, and Chris Berg who helped with an earlier version.
Any errors, omissions, or (gasp!) unpopular opinions are my own.
There are a variety of references to Google products in this document. To provide more context, I give a short description of the most common examples below.
YouTube is a streaming video service. Both YouTube Watch Next and YouTube Home Page teams use ML models to rank video recommendations. Watch Next recommends videos to watch after the currently playing one, while Home Page recommends videos to users browsing the home page.
Google Play has many models solving a variety of problems. Play Search, Play Home Page Personalized Recommendations, and ‘Users Also Installed’ apps all use machine learning.
Google Plus uses machine learning in a variety of situations: ranking posts in the “stream” of posts being seen by the user, ranking “What’s Hot” posts (posts that are very popular now), ranking people you know, et cetera.