- Maxim Ziatdinov, Center for Nanophase Materials Sciences (CNMS), Oak Ridge, TN
In this seminar, I will present an experimentalist’s perspective on the applications of deep learning neural networks in structural and functional imaging of nanophase materials. Specifically, I will talk about deep learning convolutional neural networks (DLCNN), which represent one of the key examples of a successful application of neuroscientific principles to the field of machine leaning. The DLCNN-based image analysis has been successfully used in recent years in various areas of science and engineering ranging from cancer detection to satellite imaging but has yet to be applied to atomically resolved imaging. I will discuss our recent studies on the use of DLCNN towards analysis of subnanometer-resolved scanning probe and electron microscopy images of graphene. Particularly, we were able to demonstrate a DLCNN-based approach for automated classification and localization of different types of atomic defects on graphene surfaces, such as three-fold coordinated and four-fold coordinated silicon adatoms. In addition, adaptation of convolutional autoencoders for “cleaning” of the fast scan electron microscopy images allowed us to reconstruct atomic movements (“atomic dances”) around a nanohole in single graphenic layer and to analyze the associated particles dynamics at the nanohole’s edge.
Center for Nanophase Materials Sciences Seminar Series Co-Organized with the Institute for Functional Imaging of Materials