JEOL Prize for AI-driven data analysis

Automated denoising improves nano-X-ray diffraction results

At a glance

• Award: The research team received the JEOL Prize at the Machine Learning for Microscopy Hackathon 2.

• Research focus: Improving the interpretation of nano-X-ray diffraction (nano-XRD) reciprocal space maps (RSMs) by removing noise from experimental data with the help of machine learning.

• Relevance: The model was trained on around 17,000 RSM frames and can be trained in about 30 minutes on a single NVIDIA A100 GPU, enabling rapid data evaluation during experiments.

At the Machine Learning for Microscopy Hackathon 2, held on 16–18 December 2025, Dr. Navyanth Kusampudi, Soroush Motahari, and Kartik Sunil Umate received the JEOL Prize for improving the analysis of reciprocal space maps (RSMs) from scanning nano-X-ray diffraction (nano-XRD) experiments using machine learning.

What the team developed

The team applied a self-supervised Noise2Self convolutional neural network (CNN) to remove pixel-scale noise from nano-XRD data.

Unlike conventional filtering approaches, the method:

• does not require ground-truth data

• avoids manual intensity thresholds

• preserves the shape of diffraction peaks

Why it matters

The approach significantly improves peak visibility and enables more robust centre-of-mass (CoM) and strain-field mapping, even under low-dose or short-exposure measurement conditions.

This is especially valuable during synchrotron beamtime, where limited measurement time makes reliable analysis of low-count data essential.

Fast and practical

The model was trained on around 17,000 RSM frames and completed training in about 30 minutes on a single NVIDIA A100 GPU, making it practical for near-real-time checks during experiments.

The Max Planck team received the prize together with their collaborators, Meizhong Lyu from the University of Michigan (USA) and Leonardo Oliveira from Lund University (Sweden).

The work shows how artificial intelligence and machine learning can be used for enhancing experimental analysis.The researcher team is now working on developing an end-to-end workflow.

The hackathon was supported by several sponsors and a co-organizing team.