When: 11th of October 2022

Where: Inria Paris, 2 Rue Simone IFF, 75012 Paris

Sponsorship survey:  Link (5 minutes)


8h00-8h30 (Paris Time) Breakfast – Espace Daumesnil (Ground floor Inria Paris)
8h40-8h50 Opening words – Jacques-Louis Lions Room (Ground floor Inria Paris)
8h50-9h20 Intro Keynote – Why we need MLOps ? – Franck Charras (INRIA) – Jacques-Louis Lions Room

Abstract: MLOps is quite a new engineering discipline and only became a trending vocabulary a couple of years ago. This introductory talk tries to capture the set of needs it addresses, and how it matters within the design, development and deployment teams and workflows of machine learning embedded software. We will cover practices that are common with the more classical framing of devops, and some that are specific to data pipelines and machine learning systems

The speaker: Currently working as a machine learning software engineer at INRIA, Franck Charras co-founded and worked at Sancare as data scientist and CTO, and was lead in the development from POC to industrialization at-scale of a machine-learning embedded billing software for healthcare administrations.

9h20-10h00 Shift Happens – Léo Dreyfus-Schmidt (Dataiku) – Jacques-Louis Lions Room

Abstract: This two-part talk will start with a general overview of the state of MLOps at various organizations as seen by a Dataiku’s product manager. We will then change gears and present some of Dataiku’s research work underpinning the drift detection methods used at Dataiku. In particular, we will explain why it’s not enough to detect data drift but rather try to estimate model performance drop.

The speaker: Léo Dreyfus-Schmidt is VP Research at Dataiku. He is responsible for the definition and execution of the company research roadmap.

10h00-10h40 ML Ops is first and foremost a human adventure – Guillaume Chervet (AXA)

The speaker: Guillaume Chervet  has experience putting in production Machine learning models in the insurance sector. He is Tech Lead and Machine Learning Engineer @AXA France

10h40-11h00 Break – Espace Daumesnil
11h-11h30 Document your ML models with Model Cards – Merve Noyan (Hugging Face) – Jacques-Louis Lions Room

Abstract: Versioning machine learning models require different practices compared to versioning your code. This includes documentation of the model. Model cards help us document models for the sake of reproducibility and transparency to enhance MLOps practices. In this talk, I will walk you through model cards, the tools we’ve developed to programmatically create them and will provide a small hands on demonstration on how you can create your model card.

The speaker: Merve Noyan is an open-sourceress at Hugging Face. She is a developer advocacy engineer and a developer of Skops.

11h30-12h00 Interoperate Hugging Face with scikit-learn: Skops – Adrin Jalali (Hugging Face) – Jacques-Louis Lions Room

Abstract: At Hugging Face, we are working on tackling various problems in open-source machine learning, including, hosting models securely and openly, enabling reproducibility, explainability and collaboration. We are thrilled to introduce you to our new library: Skops! With Skops, you can host your scikit-learn models on the Hugging Face Hub, create model cards for model documentation and collaborate with others.

The speaker: Adin Jalali is a computer scientist / bioinformatician who became a core developer of scikit-learn and fairlearn, and work as a Machine Learning Engineer at Hugging Face.

12h00-14h Lunch break @The Godfather restaurant, 41 All. Vivaldi, 75012 Paris

Check the menu: https://bit.ly/3RQsav6

14h-18h30 Technical Committee @Inria Paris

With the scikit-learn consortium team and the representatives of sponsors.

14h-15h – Presentation of the technical achievements and ongoing work by Olivier Grisel

15h16h30 – Feedback and exposition of each partner of the consortium


16h30-17h – Break

17h-18h30 – Collaborative drafting session of the updated roadmap