Criteolabs at Berlin Buzzwords 2015

When: Sunday 31 May – Wed 3 June, 2015
Location: Berlin

Berlin Buzzwords is Germany’s most exciting conference on storing, processing and searching large amounts of digital data. It focuses on open source software projects. Berlin Buzzwords is known for its eclectic program: Presentations range from beginner friendly introductions on hot data analysis topics to in-depth technical presentations about scalable architectures. Read more here:  http://berlinbuzzwords.de/

Speakers from Criteolabs:

Bio

Simon Dollé joined Criteo in early 2015 as a Senior Software Engineer.
He is working on the recommendation system that chooses which products to display on the ad banners.Previously he was Technical Lead at LTU Technologies, developing a large scale image fingerprinting system. Simon holds M.Sc. in Computer Science from the Ecole Centrale Paris(France) and a M.Sc. with a major in intelligent systems from Technische Universität Dresden (Germany).

Abstract

Recommendations at Scale

Behavioral retargeting consists of displaying online advertisements that are personalized according to each user’s browsing history. At this point, the selection of the products to display in the banner needs to be fast and accurate. At Criteo, we built a recommender system which is able to choose a dozen of relevant products from over two billion products in a few milliseconds. In this talk, we will expose the problems we faced whilst building this system and how we solved them thanks to a mix of online and offline computations.

Olivier Toromanoff

Bio

Olivier is Senior Software Developer at Criteo and hold as Master of Science in Computer graphics and digital image processing from Technische Universität Wien in Viena, Austria.

Olivier is based in Paris and he is  fluent in French, English, Dutch and German.

Abstract

Prediction @ Criteo: learning at scale on Hadoop

Machine Learning is at the heart of the Criteo platform : Sourcing the inventory and generating the banners for 2 billion daily displays require 15 million predictions per second. In this talk, we will present the infrastructure that allow us to train and deploy 1700 models a day as well as the tooling we set up to continuously monitor and improve this pipeline.