Machine Learning algorithms build a mathematical model based upon representative sample data, or ‘training data’, to make predictions or decisions without being explicitly programmed to perform the task. How might this be applied to supervised learning in the context of seismic data image processing of a real marine seismic dataset? How does the computational scale of such an exercise reinforce the need to develop computing technology that is customized for large machine learning problems?
In this article, Andrew Long looks at these issues and then briefly describes the emergence and relevance of application-specific integrated circuit (ASIC) concepts to accelerate machine learning applications.
Left: before | Right: after | The coherent noise on this migrated stack has been reduced after footprint attenuation using a convolutional neural network (CNN) without significant damage to the image resolution
Contact a PGS expert
If you have questions related to our business please send us an email.