Artificial Intelligence meets Nanoscience with Andy Sode Anker

By Pranoti Kshirsagar /

Andy Sode Anker’s group makes new functional materials, takes them to synchrotrons to follow the chemistry with X-rays as it happens, and ends up with a lot of data. Analysing that data with artificial intelligence is, in his words, the core of his research. He moved from the lab bench to data after deciding he was too clumsy and not patient enough for lab work.

About the guest

Andy Sode Anker was a PhD student in materials chemistry at the University of Copenhagen when we recorded this episode in late 2022. He grew up on the Danish island of Bornholm and worked closely with the university’s Department of Computer Science. He has since moved to the Technical University of Denmark, where he continues to develop machine learning methods for analysing X-ray scattering data from nanomaterials.

Andy Sode Anker curated the Real Scientists Nano Twitter account (@RealSci_Nano) from November 28 to December 4 2022. Read all of Andy’s tweets from that week.

Key highlights in this episode

  • 03:51 Chemistry meets AI. His group makes new functional nanomaterials, follows their formation with synchrotron X-rays, and then uses artificial intelligence to analyse the huge datasets.
  • 06:50 Too clumsy for the lab. Starting in the lab in his first bachelor’s year, he moved to synchrotron experiments. His first project gave him 50,000 to 100,000 datasets, which pushed him into programming and AI.
  • 13:12 From 3,000 dimensions to two. AI compressed data on chemical structures into a simple two-dimensional map, showing how stacking faults relate to the FCC and HCP crystal structures.
  • 20:46 Talk, then walk. He needs both team discussions and time alone. A 20-minute walk to think about a problem is part of the job.
  • 23:43 Three wishes. Big-tech-scale computing power, his own research group, and a much faster loop from idea to experiment to result.
Artificial Intelligence meets Nanoscience with Andy Sode Anker
Listen on

Why it matters

A single synchrotron project gave Andy between 50,000 and 100,000 datasets, which takes a long time to analyse by hand. AI made it possible to compress data with 3,000 dimensions into a two-dimensional map, so similarities between structures become visible. Defects such as stacking faults, normally just labelled and not well understood, can be placed and compared. Andy stresses that nobody can do this kind of research alone: it takes people who know the applications, the synchrotron experiments and AI. It also changes the pace. Synchrotron experiments can take a year from proposal to beam time, while work with simulated data can move much faster.

Terms explained

In the guest’s own words, with a link to the moment they explain it.

  • Synchrotron: a big ring with electrons moving around, used to generate X-rays that are shot into a sample to follow a chemical reaction while it happens. (Andy at 03:51)
  • FCC and HCP: two crystal structures. HCP (hexagonal close-packed) stacks its layers ABAB; FCC (face-centred cubic) stacks them ABCABC. (Andy at 14:40)
  • Stacking fault: a layer sequence that is neither FCC nor HCP, for example ABA followed by C. (Andy at 15:12)
Listen on

Part of AI, Modelling & Theory · AI & machine learning

All AI, Modelling & Theory episodes →