MARVEL-wide Metatensor Ecosystem Workshop

Request edit access

JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

Dear MARVEL community,

It is with great pleasure we are announcing the organization of a MARVEL-wide "Metatensor Ecosystem Workshop", to be held on June 13th 2025 @ EPFL main campus.

Our workshop introduces the Metatensor ecosystem, an interoperable collection of software from COSMO developed in part for MARVEL Pillar 2.

Metatensor addresses all the practical needs associated with atomistic machine learning (ML), from data storage and handling to model training with various ML architectures for a variety of learning targets.

Our day-long workshop will first provide a general introduction to Metatensor and its related libraries in the morning, then break off into two parallel sessions in the afternoon: (1) a deeper dive into metatensor for ML model developers, and (2) how to train your model -- training custom models on custom datasets, and (3) fine-tuning the PET-MAD universal model with metatensor for specific applications.

While the event primarily targets past and present members of the MARVEL community, we are open to welcoming other researchers from the wider research community interested in attending (conditional on the remaining capacity).

If you are interested, please sign up below. We look forward to welcoming you in Lausanne! :)

Sincerely,

Your Metatensor Ecosystem Workshop organizers

(Please direct any queries to Sanggyu "Raymond" Chong <sanggyu.chong@epfl.ch> and Guillaume Fraux <guillaume.fraux@epfl.ch>)

First Name *

Last Name *

Email *

MARVEL affiliation *

Pillar 1 - Novel Materials

Pillar 2 - Machine Learning

Pillar 3 - Open Digital Infrastructure

Pillar 4 - Swiss Scientific Landscape

Advanced Simulation Methods

Quantum Simulations

I'm from a group with past MARVEL affiliation

I have no MARVEL affiliation, but I would like to attend!

After the introductory morning session, which parallel afternoon session would you like to attend?

In case too few people are interested, we might cancel one of the sessions. *

Diving into metatensor and metatomic architecture for ML model developers

How to train your model (training custom models on custom datasets)

Fine-tuning interatomic potentials to improve accuracy on your own data

Submit

Clear form

Never submit passwords through Google Forms.

This form was created inside of EPFL.

Does this form look suspicious? Report

Forms