DL in Production

In this event, you will be hearing about the journey to deploying the first deep learning pipeline into production for audience-targeted advertising. Maria and Sofie discuss the switch in the technical stack from Spark MLlib to Tensorflow, touching upon the different stages that comprise the new pipeline. They zoom in on the transformations of raw events by going sparse, and the different approaches for handling high dimensional text features. They also talk about reformulating the prediction problem from a tree-based algorithm, XGBoost to a WaveNet-like sequence model and their dreams about what’s next. 00:00:00 Stockholm AI Introduction - Ingrid af Sandeberg 00:04:24 AI @ Schibsted - Armin Catovic, Schibsted 00:10:35 Enabling Deep Learning at scale by going Sparse - Maria Slanova & Sofie Verrewaere, Schibsted 00:42:35 Q&A Session