AI Dev Talks is a meeting place for engineers, researchers and students who want to share practices, tools and / or advice around technical topics related to Artificial Intelligence applications.

For this new session, two new presentations are on the agenda:

- Introduction to the use of MLFlow (Ghisalin Vaillant - CNRS / Institut du Cerveau)

 The MLFlow open source platform enables the lifecycle management of machine learning experiences, ensure its reproducibility, deploy the results and manage the management of artifacts (models) from these experiments in a simple and intuitive way. This presentation explores the concepts of basis for using this tool.

- Geomstats: Development of a geometry library for statistical learning (Thomas Gerald, Nicolas Guigui - Sorbonne Université, Inria)

In this presentation, we will talk about the Geomstats library: one Riemannian geometry library for statistical learning.
We will first present the main principles of the package as well several application cases. In a second step, we will discuss his  development: the organization of sprints, the practices set up and the various tools used for its deployment and maintenance of the code. With the rise of applications and deep learning libraries, we will discuss choices for package interoperability with leading backends: Numpy, PyTorch and Tensorflow. Finally, we will discuss the steps to be taken for the integration of news features such as matrix computing unit support (GPUs).