XAI Generated Blind-masks for Self-Supervised Seismic Denoising: Claire Birnie (KAUST) скачать в хорошем качестве

XAI Generated Blind-masks for Self-Supervised Seismic Denoising: Claire Birnie (KAUST)

Dr. Claire Birnie, research scientist at King Abdullah University of Science and Technology, gave a presentation titled "XAI Generated Blind-masks for Self-Supervised Seismic Denoising" (1 Feb. 2024) Abstract: Self-supervised denoising overcomes the challenge posed by traditional deep learning's requirement of clean training labels - an unobtainable requirement for field seismic data. I will begin by introducing blind-spot and blind-mask denoising, illustrating their application to field seismic datasets. For coherent noise suppression, a mask on the input to the network must be defined as part of the training process associated with a self-supervised denoising network. Such a mask however requires prior knowledge of the properties of the contaminating noise, often assuming that noise can be sampled without signal. The second part of the presentation will focus on how by incorporating explainable AI techniques in blind-spot denoising, namely Jacobian inspection, we can automatically identify the noise characteristics and design an optimal noise mask, without requiring any signal-noise separation. The presentation will conclude with field data examples of the automated XAI-denoising workflow.