Transmission electron microscopy (TEM) offers a wide range of techniques by which the atomic structure of matter can be probed, making use of elastic and inelastic interactions of fast electrons with the probed atomic structure and the distribution of electrons within it. Starting from a known atomic structure, it is now possible to simulate many of the experimental observations we would obtain from it with high accuracy. However, solving the inverse problem to retrieve the atomic structure from the experimental data is still a major challenge and works only in very few cases.
In this talk I will first introduce the potential that TEM offers, present the current state-of-the-art in (big) data analytics in TEM, and introduce our own plans in this direction. I will also show how we are using special multi-layer artificial neural networks to solve the inversion problem of multiple elastic electron scattering, in principle recovering three-dimensional atom positions from TEM data. During the talk a number of open questions in line with the critical open questions defining the scope of this workshop will be presented.
Back to Workshop II: HPC and Data Science for Scientific Discovery