Sep 2023
Deep learning has revolutionized almost every field of science. However, traditional deep learning procedures are often supervised requiring a ground truth label that is used to train the network. In the field of geophysics, this label is often unobtainable, with the ground truth remaining unknown. One approach to still leverage deep learning in geophysics is to implement self-supervised methods, where the available data represents both the input and label for training a neural network. Given the example of noise suppression, in this presentation, I will illustrate how blind-spot, and subsequent blind-mask, networks can be utilised for the suppression of both random and coherent noise. Furthermore, I will introduce how we can exploit methods from the field of explainable artificial intelligence in order to design the optimum blind-mask, resulting in tailored noise suppression algorithms that require no clean training targets.
Claire is a research scientist in KAUST developing new methods/techniques at the intersection between geophysics and deep learning. She holds a PhD in computational geophysics from the University of Leeds and was awarded a Microsoft scholarship to undertake their 6-month professional data science certification. Previously a senior data scientist in Equinor’s (nee Statoil) Digital Center of Excellence, in KAUST she combines her industry experience with her technical know-how to develop research solutions with an immediate industry application.