SurfaceScience21 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 55: Poster Session IV: Poster to Mini-Symposium: Machine learning applications in surface science II
O 55.5: Poster
Dienstag, 2. März 2021, 13:30–15:30, P
Image-to-graph translation of atomic force microscopy images using graph neural networks — •Niko Oinonen1, Fedor Urtev1,2, Alexander Ilin2, Juho Kannala2, and Adam Foster1,3,4 — 1Department of Applied Physics, Aalto University, Finland — 2Department of Computer Science, Aalto University, Finland — 3Graduate School Materials Science in Mainz, Germany — 4WPI Nano Life Science Institute, Kanazawa University, Japan
The atomic force microscope (AFM) is an important tool in nanoscale science for imaging surfaces and molecules on surfaces. State-of-the-art AFM setups operating in vacuum at low temperatures are able to resolve features on the scale of individual atoms in molecules. However, the process of interpreting the resulting AFM images in some cases can be very challenging even for highly trained experts in the field. We are working towards greater interpretability and greater automation of the processing of AFM images using machine learning methods [1]. We are currently exploring the possibility of directly predicting the atomic structure of the sample as a graph using graph neural networks (GNN) [2]. We propose a GNN model which, conditioned on an AFM image, iteratively constructs the graph of the sample molecule present in the AFM image, following similar work by Li et al. [3]. This is still a work-in-progress, but our initial results are showing promise.
[1] B. Alldritt et al. Sci. Adv. 6(9), eaay6913, 2020.
[2] P. W. Battaglia et al. arXiv:1806.01261.
[3] Y. Li et al. arXiv:1803.03324.