OceaniX members organize different courses and training sessions on deep learning, data science, ocean remote sensing. We provide below more information on the following courses:

  • Doctoral course on Data Science for Geosience link
  • Doctoral course on Deep Leearning and Geophysical Dynamics link

Doctoral course on Data Science for Geoscience

This doctoral course co-organized between Grenoble, Toulouse and Brest is an introductory course to data science for geoscientists. It covers generic theoretical concepts in data science and machine learning as well as various approaches and problems. Besides labsession using python, it also involves a group project to apply these concepts to a specific issue from the formulation of the problem to the interpretation of numerical experiments.

More information can be found from the website of last session held in Januatry 2020 link. We plan to organize next session in Brest in January 2022. Do not hesitate to contact us for any additional information.

Advanced Course on Deep Learning and Geophysical Dynamics

Next october we will organize the first edition of an advanced course on Deep Learning and Geophysical Dynamics co-organized by AI chairs OceaniX, DL4CLIM, ANITI-DAML and AI4Child (Prof. R. Fablet, P. Gallinari, S. Gratton and F. Rousseau). The general objective of the course will be to cover theoretical aspects of deep learning and its application to geophysical dynamics, especially regarding the exploitation of physical priors. Besides these theoretical aspects, the course will also aim to apply state-of-the-art deep leearning approaches to specific case-studies using pytorch.

The course will take place on tuesdays five consecutive weeks from November 9 to December 7. All morning sessions (typically, 9.30am-12.30pm, Paris time) will be dedicated to a course covering theoretical aspects combined with some pytorch practice, while all afternoon sessions will be dedicated to an advanced practice through a specific team project.

The provisional program for the different morning session is the following:

  • Course #1 (Nov. 9): Introduction to Deep Learning and Automatic Differentiation (Prof. F. Rousseau)
  • Course #2 (Nov. 16): Deep Learning and Optimization (Prof. L. Drumetz and Prof. Serge Gratton)
  • Course #3 (Nov. 23): Generative models (Prof. P. Gallinari) NB: this course will be held from 2.30pm and the morning session will be dedicated to the project
  • Course #4 (Nov. 30): Deep Learning and Dynamical Systems (Dr. S. Ouala)
  • Course #5 (Dec. 7): Deep Learning and Inverse Problems (Prof. R. Fablet)

The introductory project session on November 9 will also include a short introduction to pytorch and Pytorch Lightning which will be the reference deep learning framework of the course. Examples of projects of interest for geophysical dynamics (more to come):

  • Space-time interpolation of sea surface dynamics
  • Forecasting of El Niño index
  • Learning numerical schemes for ODEs and PDEs
  • Data-driven system identfication and forecasting

Pre-requisite: Participants shall have some basic knowledge in applied math. and/or machine learning and some practice in Python programming. Previous knowledge in deep learning and pytorch is not a pre-requisite. We will provide some links prior to the course to online tutorials for Pytorch, so that all participants could gain at a first basic practice before the course.

Online resources to prepare the course For those who have no or very little knowledge on deep learning and pytorch, below some resources:

  • Introductory course from Stanford Univ. on deep learning video
  • Reference book on deep learning link
  • Introduction to pytorch framework link
  • 60-minute blitz for pytorch link

Organization: The course will be organized in person in Brest at the “Pôle Numérique Brest Iroise”. It will also be possible to attend to the courses remotely. We expect to accomodate up to 50 participants for the course.

Registration deadline: Deadline passed (contact ronan.fablet AT imt-atlantique.fr for additional information)