Much of the seismic data analysis has been carried out using methods based on the physics of wave propagation and some signal processing principles. These methods are generally termed 'physics-based' or 'model-based' approaches. The entire suite of seismic processing and inversion algorithms starting from stacking velocity analysis to full waveform inversion is developed on these principles. Recently, a flurry of data-driven models that are agnostic to the physics has been developed for application to some previously solved and unsolved problems. Physics based approached rely on fundamental physical principles resulting in some predictive equations. However, it may suffer from the limitations of being incomplete. Data driven approaches, on the other hand, are methods for understanding the mechanisms generally by identifying patterns in large volumes of data. Thus, it may appear that we are at cross-roads to decide on the applicability and usefulness of the two seemingly different approaches. Even a broader question is "do we need to choose between the two?"

Questions answered during this course:
· How do we relate input data to the desired output?
· What are the appropriate questions to ask?
· What is the role of physics in seismic data analysis? A historical perspective.
· Are we missing something in our physics-based models? What are the limitations, if any?
· Can the signal processing-based approaches address some of the limitations?
· What is a data-driven or a machine learning (ML) approach?
· How do ML methods work? Is ML just the old wine in a new bottle?
· What is the role of physics in ML formulation?
· Do ML methods offer any advantages over current physics-based approaches?