ABSTRACT
In this chapter, the authors delve into the fundamental concepts behind scanning tunneling spectroscopy and highlight machine learning methodologies that employ, e.g., neural networks and GPs to drive experimentation. Gaussian process regression has enabled a means for autonomous hyperspectral data collection on relevant 2D van der Waal materials, and can be subsequently classified by convolutional neural networks after a successful experiment which can be applied to any surface that is characterized by a scanning probe microscope. The strategies to overcome this challenge specific to the data sets typical to scanning tunneling microscopy are discussed. In 2021, Wang et al. demonstrated optimization for spectroscopy resolution by classification of Au STS spectra until the surface state was visible. Classification results obtained from applying a VGG-like convolutional neural network to recognize features commonly observed in images of Au during the tip conditioning and evaluation process are presented.
