ParAMS Machine Learning Challenge

We are pleased to announce the ParAMS Challenge for 2025 to coincide with the WATOC 2025 conference, which SCM is attending!

We invite you to test your machine learning and simulation skills by fine-tuning the M3GNet machine learning potential using ParAMS. 

Goal: improve the performance of M3GNet for polymerization reactions

Prize: best entry receives 250 EUR Amazon voucher or equivalent

Eligibility: all attendees of WATOC 2025. Sign up at the SCM booth!

Expected time investment: 1+ hours of your time depending on experience. 20+ minutes of computer time. You can of course spend as much or as little time as you want.

Submission deadline: 15 July, 2025

Submission deliverable: send the ParAMS results directory as a zip file to [email protected], together with your name and affiliation

Questions: send any questions about the challenge to [email protected], or ask at the booth

Tip: Before attempting the challenge, check out the Getting Started, Train M3GNet with the ParAMS GUI, and Train M3GNet with python tutorials

How to Participate

Follow these steps to get started with the challenge:

• Provide your details at the SCM booth at WATOC 2025 to get a trial username and password
• Download and install AMS2025
• Install M3GNet using AMSpackages
• Download and unzip params_ml_challenge_2025.zip
• Run the provided example (~20 minutes on a laptop) via the GUI or command line (see the README)
• Improve the forcefield by:
  • Modifying training parameters
  • Adding additional training and validation data
  • Using molecular dynamics and active learning strategies
  • Any other methods you can think of!

Once you have your best solution, submit your entry to [email protected], by 15 July 2025.

Rules

SCM provides a training and validation set, that you will use during the parametrization. However, the goal of the challenge is to get the smallest error on the test set, that will only be made public after the challenge ends.

Only one submission is permitted per person.

If you win the challenge, SCM would like to post an interview with you on its website.

SCM may choose to extend the duration of the challenge, and may disqualify an entry for any reason at its sole discretion.

Data Sets

The training and validation sets contain data for describing an epoxide-amine polymerization reaction.

It contains

• Conformer energies generated with AMSConformers
• Energies and forces from a bond boost trajectory (reactive events identified by ChemTraYzer2)
• Bond scans

The test set contains reaction energies, optimized geometries and data similar to the training and validation sets, for similar types of molecules.

All training and test data was calculated using the PBE functional with D3(BJ) dispersion corrections and a TZP basis set.