Seismic attribute calculations have been traditionally used for measuring specific traits in seismic data for further interpretation (Chopra and Marfurt 2007). These calculations tend to have a fixed operator or window size from one sample to the other during the computation. They also tend to focus on single measurements and rely heavily on user input for parameterization. This tends to make it hard to reproduce results between users or even understand the certain choices made by various experts as they do their measurements and calculations on the seismic data. This can also be very limiting given the varied nature of the seismic signal and aspects such as seismic attenuation. One will tend to easily over sample or under sample at any given state in the seismic data as the signal changes, ultimately yielding results that need to be tweaked repetitively to get the perfect fit for the given depth/time the features of interest are in. We have looked at how we can calculate signal decomposition features for typical fluvial system detection in an adaptive fashion to overcome some of these challenges (Aqrawi and Aqrawi 2014).
In the past few decades, the concept of using Artificial Intelligence (AI) to develop Machine Learning (ML) models to solve challenging issues in the various industrial sectors and in academia has skyrocketed, and for a good reason as well. As time has gone on, it has become clear that the "fourth industrial revolution" is the digital transformation, which is characterized by the convergence of technologies like robotics, artificial intelligence, and autonomous vehicles that blur the lines between the physical, digital, and biological realms (Sircar, et al. 2021) (See Figure 1, which represents the various fields of AI and ML).
Machine learning, which is a subset of AI, is defined as a computing paradigm in which the ability to solve the presented problem is developed by referring to similar prior cases called training examples. It reasons the problem by "learning" from the training examples. There are several tasks to which we can apply ML methods to. These include:
