Sixth FAIRmat Users Meeting

Jul 2, 2025, 9:00 AM → Jul 3, 2025, 4:00 PM Europe/Berlin

UFO (Ruhr-Universität Bochum)

We are pleased to announce the Sixth FAIRmat Users Meeting, taking place at Ruhr-Universität Bochum on July 2-3, 2025.

The FAIRmat Users Meeting is a central event for scientists in physics, chemistry, and materials science who are passionate about research data management (RDM) and its role in accelerating scientific discovery.

The upcoming meeting will focus on AI- and ML-driven materials science and how the FAIR data principles make research data "AI-ready". Learn how NOMAD, our free web-based platform, is more than just a data repository – it is an enabler for next-generation materials discovery using machine learning and artificial intelligence.

Whether you are a seasoned NOMAD user, exploring its potential, or just starting your RDM journey, this event will offer valuable insights, practical tools, and meaningful connections.

What to expect:

• Invited talks from experts using NOMAD and NOMAD Oasis for AI/ML applications in computation, experimentation, and synthesis.
• Hands-on tutorials on the NOMAD AI Toolkit, downloading ML-ready datasets, and getting started with NOMAD.
• Poster session open to external contributions.
• Informal support desks for direct interaction with NOMAD developers and FAIRmat domain experts.
• Networking opportunities.

We look forward to two days of stimulating discussions, hands-on sessions, and community building!

Wednesday, July 2

9:00 AM → 10:30 AM Harnessing Data-Driven Methods in Materials Discovery

9:00 AM FAIR Research Data Management with FAIRmat and NOMAD

In this talk, I will present the latest developments of FAIRmat and the NOMAD ecosystem, emphasizing the critical aspects of FAIR research data management (RDM) in condensed-matter physics. The NOMAD ecosystem, which includes platforms such as NOMAD, NOMAD Oasis, and NOMAD CAMELS, provides a complete solution for an RDM framework tailored to the needs of the condensed-matter physics and materials-science communities. I will demonstrate how these platforms facilitate the systematic acquisition and management of research data from diverse sources, including materials synthesis, advanced characterization techniques, and simulations. The presentation will also cover how NOMAD ensures that research data is not only securely stored, but also remains accessible and interoperable across different research domains. This in turn supports collaborative research efforts and accelerates innovation by ensuring that data is AI-ready.

Speaker: Hampus Näsström (Humboldt-Universität zu Berlin, FAIRmat)

9:30 AM Quantitative XPS Analysis Using Convolutional Neural Networks

X-ray photoelectron spectroscopy (XPS) is a powerful tool for studying the electronic structure and chemical composition of solid surfaces. Quantitative analysis of XP spectra typically relies on manual curve fitting by expert spectroscopists. However, recent advancements in the ease of use and reliability of XPS instruments have led to a growing number of (novice) users generating large datasets that are becoming difficult to analyze manually. Additionally, the expansion of publicly available XPS databases further increases the volume of data requiring efficient analysis. Reflecting these developments, more automated techniques are desirable to assist users in processing large XPS datasets.

Here we present a scalable framework for automated XPS quantification using convolutional neural networks (CNNs). By training CNN models on artificially generated XP spectra with known quantifications (i.e., for each spectrum, the concentration of each chemical species is known), it is possible to obtain models for auto-quantification of transition metal XP spectra [1]. CNNs are shown to be capable of quantitatively determining the presence of metallic and oxide phases, achieving accuracy comparable to or exceeding that of conventional data analysis methods. The models are flexible enough to handle spectra containing multiple chemical elements and acquired under varying experimental conditions. The use of dropout variational inference for the determination of quantification uncertainty is discussed. Finally, we demonstrate how these network models are integrated into NOMAD [2], enabling real-time analysis of newly generated data.

References
[1] Pielsticker, L.; Nicholls, R.; DeBeer, S.; Greiner, M., Analytica Chimica Acta 2023 1271, 341433.
[2] Scheidgen, M., et al., Journal of Open Source Software 2023 8.90, 5388.

Lukas Pielsticker [1], [2], Rachel L. Nicholls [1], Serena DeBeer [1], Walid Hetaba [1], Florian Dobener [2], Laurenz Rettig [3], José A. Márquez [2], Sandor Brockhauser [2], Heiko Weber [4], Claudia Draxl [2], Mark Greiner [1]

[1] Max Planck Institute for Chemical Energy Conversion, Mülheim an der Ruhr, Germany
[2] Physics Department and CSMB, Humboldt-Universität zu Berlin, Germany
[3] Department of Physical Chemistry, Fritz Haber Institute of the Max Planck Society, Berlin, Germany
[4] Chair of Applied Physics, FAU Erlangen-Nürnberg, Erlangen, Germany

Speaker: Lukas Pielsticker (MPI for Chemical Energy Conversion)

10:00 AM Small-Data Models for Materials Design

The modeling of macroscopic properties of materials often require to accurately evaluate physical quantities at several time and length scales. Here we show how symbolic inference, i.e., the machine learning of simple analytical expressions that explain and generalize the available data, can effectively bridge physical scales. The focus is on learning models that are as simple as possible (but not simpler…), with as few as possible data points. I will demonstrate the application of the methods to the modeling of catalytic properties of materials, thermal conductivity, and more.

Speaker: Luca Ghiringhelli

10:30 AM → 11:00 AM Coffee Break

11:00 AM → 12:00 PM Exploring AI & ML in Simulations, Databases, and Computational Workflows

11:00 AM The Alexandria Database of Materials

The Alexandria database represents an important resource for materials science research, containing more than 5 million density-functional theory calculations for periodic three-, two-, and one-dimensional compounds. This open-access database addresses the critical challenge of data scarcity in materials science, where large, high-quality, and consistent datasets are rare. The Alexandria database serves as a foundation for training machine learning models in materials science, enabling the training of multiple material properties using both composition-based models, crystal-graph neural networks, or machine-learning interatomic potentials. This is also an important resource in accelerating materials discovery and in transforming high-throughput materials design. The Alexandria database exemplifies the power of open science in materials research, providing the community with a robust platform for developing predictive models and discovering new materials with tailored properties.

Speaker: Miguel Marques

11:30 AM Creating Data Analysis Pipelines Using the MADAS Framework

The NOMAD data infrastructure provides access to vast amounts of data that can be used for data analytics and machine learning (ML). Often, however, not all (meta)data are relevant for every task, making it necessary to apply filtering and processing steps to prepare input data for ML.
Here, we present MADAS, a Python framework that supports all steps of data analytics and machine learning, including automated download and storage of data, generation of material descriptors, and computing similarity metrics, and integrates well with established ML frameworks and libraries. MADAS allows to write robust, re-usable data analysis pipelines, while its modular structure allows to quickly extend the data processing with custom functions.
We demonstrate its capabilities and features by finding interoperable data within a large computational dataset hosted on NOMAD, and by finding distinct materials that exhibit similar electronic structures.

Speaker: Martin Kuban

12:00 PM → 1:00 PM Lunch

1:00 PM → 3:00 PM Hands-on Workshop

1:00 PM Role-Based NOMAD Usage and Development

NOMAD now provides a broad ecosystem of data infrastructure software and tools, enabling robust data management at the individual, research group, and institutional level. To navigate this ecosystem, it is useful to clearly identify your role and desired usage. In this tutorial, we will assist you in getting started down the right path, whether you are already an experienced user or are brand new to NOMAD.

You will choose from one of the following roles/topics to explore NOMAD’s capabilities:

USER
• Project and workflow management (recommended for users with a basic Python setup and minimal coding knowledge): organize, process, share, and publish datasets; interact programmatically via a simple Python API; customize entries and workflows.
• Basic NOMAD usage via the GUI: upload, share, and publish data; use NOMAD’s electronic laboratory notebook (ELN) interface with built-in schemas; create ELNs with custom extensions.

APPLICATION ADMINISTRATOR (ADVANCED USER)
• Plugin development: create a plugin repository using a cookie-cutter template; transform custom YAML/JSON schemas into Python code; add automation and plotting features.

SYSTEM ADMINISTRATOR
• Local infrastructure setup: install and configure your own NOMAD Oasis; set up CI pipelines and create custom images with plugins.

Speakers: Joseph Rudzinski (Humboldt University), Hampus Näsström (Humboldt-Universität zu Berlin, FAIRmat)

3:00 PM → 3:30 PM Coffee Break

3:30 PM → 5:30 PM Poster Session & Meet Our Experts

7:00 PM → 9:00 PM Dinner

Thursday, July 3

9:00 AM → 10:30 AM AI & ML Activities Across NFDI Consortia

10:30 AM → 11:00 AM Coffee Break

11:00 AM → 12:00 PM Language Models for Materials Science

11:00 AM Text Mining in Materials Science

Lei Zhang, Doaa Mohamed, Sepideh Baghaee Ravari, Markus Stricker

Beyond the direct raw data sources experiments and simulations, scientific publication are an underused resource at scale. The content of scientific publications can be converted into high-dimensional vector representations to gain access to the underlying correlations. Raw text can be converted to word embeddings (word2vec) and combined vision-language models can be used to extract structured datasets from scientific publications. These high-dimensional representations can then be used for data mining. I will demonstrate the potential of text mining using two examples: (1) how correlations in word embedding space can accelerate active learning loops in materials discovery, and (2) workflows for converting unstructured scientific publications to structured. However, robust and standardized pipelines for these methods are still a work in progress but these result, among others, already demonstrate useful applications.

Speaker: Markus Stricker

11:30 AM Transforming Materials Science with Transformers

The advent of large-scale transformer models has opened new frontiers in scientific discovery, with materials science poised for a significant transformation. These models promise to accelerate research by automating tasks from literature review to property prediction. This talk will provide a critical evaluation of the current state of transformer-based models in chemistry and materials science. This talk will synthesize these findings to argue that the path to transforming materials science lies not with a single monolithic model, but in a hybrid approach, where LLMs would be used to automate research task coupling with specialized models or experiments.

Speaker: Nawaf Alampara

12:00 PM → 1:00 PM Lunch

1:00 PM → 4:00 PM Hands-on Workshop

1:00 PM Large Language Models for Scientific Data Extraction

Structured data is key to machine learning – but much of the world’s information is unstructured. This tutorial introduces how Large Language Models (LLMs) can be used to extract structured data from scientific publications. After a 30-minute introduction talk, participants will dive into a hands-on session exploring practical steps of the extraction workflow.

Speakers: Mara Schilling-Wilhelmi, Sharat Patil