ML Bootcamp at Criteo

Machine Learning (ML) has been at the heart of the Criteo business since the company first began. It is the technology which allows us to present the right ad, to the right person, at the right moment. Today, ML is an important subject in any computer science course but this was not always the case.  Criteo has a need for engineers knowledgeable in Machine Learning to complement our team of specialized researchers (the Research Team) so, we came up with the idea of creating the Criteo ML Bootcamp.

What is ML-Bootcamp?

ML Bootcamp is a three-month program with the objective of immersing participants in Criteo’s Machine Learning research projects by combining both theoretical and practical training with the opportunity to contribute to a live project.

The introduction process  

To ensure that participants get the most out of Bootcamp they are required to have worked at least one year at Criteo R&D and exhibit a high level of autonomy in their work. Each applicant must submit a CV and a letter of motivation.  Each intake group is kept small to ensure the program remains highly interactive.

An intensive theoretical and practical training

The Bootcamp is composed of two complementary phases: a training phase and a mission phase. The first month is dedicated to training.  The theoretical part of the training consists of sessions covering the ten most relevant areas of machine learning delivered by members of the Research Team.  The topics covered are:

Participants are encouraged to study the content in advance and come to sessions prepared to dive deeper into each topic. The theory lessons are followed by four days of intensive, practical training sessions where the participants are given sets of exercises to perform in the different machine learning categories they’ve studied.

There are theoretical exercises, where a participant might be asked to prove the characteristics of an algorithm, as well as practical exercises, which require a participant to apply an algorithm on public data sets and study its behavior when playing with the parameters.  The idea is for participants to become comfortable using their new skills before dedicating the next two months to a project.

The mission

Equipped with a new set of tools, each participant is paired with a researcher who has a specific topic to investigate. The project includes both a scientific and an engineering component and might cover themes like:

• the investigation of different models for predicting sales based on what was observed during the first 10 hours after showing an ad to determine how we can estimate the attributed sales after 30 days;
• the study of new bounds for taking decisions on A/B tests to determine if the difference we observe on the first few days of an A/B test are already statistically significant and whether we can stop the test now;
• the implementation or porting of a policy learning algorithm in a new data pipeline.

Bootcamp missions are always based on topics the Research Team is currently investigating.  Working with a real scenario teaches engineers that putting theory into practice is not always as simple as it seems. Obtaining and cleaning data can be the most challenging part of a machine learning project. ML Bootcamp creates a collaborative situation where the Researcher makes significant progress on a project and the engineer further develops their machine learning skills.

The Debrief

After two months working on a Mission, participants present their results in a debrief session and review what lessons were learned from both an organizational and personal perspective.  Some missions have opened the door to exciting new opportunities, others have resulted in promising leads being shut down.   Regardless of the outcome, participants package up their results along with all the developed code so that it is available for future use or further development.

The adventure

ML Bootcamp is an intensive yet worthwhile program for the participants, trainers and mission leaders alike.

As with any program of its kind, the reward is directly linked to the investment that participants are prepared to make.  In order to ensure the experience remains motivating and challenging, ML Bootcamp is constantly reviewed and adapted.  It’s our goal to ensure that each participant walks away from the program feeling a sense of accomplishment and pride at having “survived” ML Bootcamp.

Post written by: