As part of our outreach program, Criteo AI Lab is proud to offer to the Machine Learning community in Paris, a course on Modern Recommendation.
This workshop, free of charge, will be delivered by a team of five Senior Researchers and ML practitioners that are part of the Criteo Recommendation group.
Â
Course structure:Â The course will be over 1 day and will have 2 parts.
In the first part, we will cover current approaches for Recommendation that are based on Empirical Risk Minimization models and their associated shortcomings.
In the second part, we will discuss ways to address the aforementioned shortcomings and introduce Policy-based methods as a potential solution.
All of the concepts will be followed by examples and coding sessions where we will put in practice the newly introduced conceptual tools.
Â
Audience:Â Daily practitioners of Machine Learning either for academic or industrial purposes that are especially active in the field of Recommender Systems.
We also encourage our attendees to be an active member in the Paris ML community.
Dates:Â December 6th, 2019
Â
Time:Â From 09:00 AM to 5:00 PM with lunch break - Food providedÂ
Â
Duration:Â 1 day
 Â
Deliverables:Â Printed slides presentations, Google Colab NotebooksÂ
Language:Â EnglishÂ
Â
Location:Â Criteo - 32, rue Blanche - 75009 Paris
 Â
Price:Â Free
Workshop attendance is by application only. Seats are limited. We will choose the participants through the application process detailed below.
Participation to the workshop will be determined based on level of proficiency in Machine Learning. You will be asked to submit your resume or LinkedIn profile when you apply, to assess the said proficiency. Also, to attend the workshop, all participants must comply with the prerequisites.
Â
Please note that the deadline for the application to the workshop is
November, Monday 18th, end of day (Paris time).
Applicants will be notified about the outcome of selection process on Friday 22nd, November, by email.
Â
I. Recommendation via maximizing likelihood approaches
1. Classic vs. Modern: Recommendation as autocomplete vs. recommendation as intervention policy
2. ERM and Likelihood models for optimal effect recommendations
3. Shortcomings of ERM/likelihood-based models for recommendation
II. Recommendation as policy learning approaches
1. Policy Learning: Concepts and Notations
2. Fixing ERM using Policy Learning
      - Fixing Covariate Shift: From ERM to Counterfactual Risk Minimization
      - Fixing Optimizer’s Curse: From ERM to Distributional Robust Optimization
      - Reco-specific Policy Learning methods: Using Organic Feedback
3. Recap and Conclusions
How to program in Python.
How to use the NumPy library.
The basics of linear algebra: Â
   - What is a vector, a matrix?
   - How to multiply and transpose them?
Know the basics of Machine Learning:
   - What is Machine Learning? Â
   - What is a model?
   - What is a cost function?Â
   - What does it mean to train a model?Â
   - What is the difference between a modelÂ
    parameter and a hyper-parameter?
   - What does "regularizing a model" mean?Â
   - What is over-fitting?Â
   - What are the training, validation, and test sets?Â
   - What is a cross-validation?
Know the basics of Recommender System:
   - What is Collaborative Filtering?
   - What is Matrix Factorization?
   - What are Precision@k and Mean Percentile Rank?