The GDR IAMAT organizes a Spring school in the beautiful location of Roscoff, focusing on machine learning in computational materials science. There will be a maximum of 45 attendants, with priority given to non-permanent researchers, and poster sessions.

The school itself will be organized on five days, from Monday 17^th of April 2023 early morning to Friday 21^st of April at noon, with, each day, two morning lectures followed, in the afternoon, by hands-on computer tutorials, which will allow participants to become operationally familiar with the machine-learning approached described and explained in the corresponding morning lectures.

The second part of two afternoons will be dedicated to poster sessions, which will be introduced orally via flash-presentations.

Confirmed speakers:

• Aurélien Decelle – Statistical methods for machine learning
• Jean-Luc Parouty – Machine learning with deep neural networks
• Jörg Behler – Neural network potentials for atomistic simulations
• Filippo Vicentini – Neural networks quantum states
• Arthur France-Lanord – Dynamics and transformations at the atomic scale using ML
• Joao Paulo Almeida de Mendonca – Learning new DFT functionals
• Ludovic Goudenège – Machine learning in phase-field approaches
• Cosmin Marinica – Overview of surrogate machine learning models for materials science
• Pauline Besserve – Beyond the ground state: accelerating DMFT with quantum computers

The normal fee of 480 euros includes housing (from Sunday 15^th of April evening to Friday morning) and meals (from Monday morning to Friday lunch). Attendance is free for participants employed by CNRS, and there will be a discount for participants employed by CEA (depending on the total number of CEA attendees).

To participate, please fill in the registration form including a few lines about your work and motivation, before the 31^st of January. Please specify in case you are directly employed by CNRS or by CEA. Applicants will be notified around mid-February whether they are selected or not.

Practical session

•  General requirements: You should bring your own laptop with docker https://www.docker.com/ and anaconda with python 3 and jupyter notebook https://www.anaconda.com/products/distribution installed. If you are not comfortable with python and jupyter notebooks, we strongly recommend that you start practicing prior to the school.

Then for each tutorial we recommend preparing a dedicated environment, with the following requirements.

• Tutorial 1 : numpy, matplotlib, scipy, h5py. The notebook can be found here: https://github.com/AurelienDecelle/RoscoffAI2023

• Tutorial 2 : the fidle environment https://fidle.cnrs.fr/installation (we recommend installation via docker or pip)

• Tutorials 3 and 5 : the docker image afldev/comp-iamat

• Tutorial 4 : anaconda and jupyter notebook. The link to the tutorial is https://github.com/PhilipVinc/Lectures/tree/main/2304_Roscoff

• Tutorial 6 : pyscf https://pyscf.org/install.html , tensorflow and ase. The link to the tutorial is https://github.com/jpalastus/Notebooks

• Tutorial 7 : tutorials can be found at https://drive.google.com/drive/folders/1vGVxd07UwA9i8ZKG-hYIxJni8sB0L2Rh?usp=sharing

• Tutorial 8 : we will provide a colab link at ai-atoms.github.io/milady

• Tutorial 9 : the link to the tutorial is https://github.com/pbesserve/Roscoff_Spring_School_tutorial
  

The slides of the lectures can be downloaded here:

Aurélien Decelle : Statistical methods for Machine Learning

Jean-Luc Parouty : ML with deep neural networks

Jörg Behler : Neural Network Potentials for Atomistic Simulations

Arthur France-Lanord : Dynamics and transformations at the atomic scale using machine learning

João Paulo Almeida de Mendonça : Artificial Neural Network-Based Exchange and Correlations Functionals

Ludovic Goudenège : Machine learning from scratch

Mihai-Cosmin Marinica : Surrogate models for atomistic materials science

Pauline Besserve : Beyond the ground state: accelerating Dynamical Mean-Field Theory with quantum computers