Abstract Machine learning (ML) pipelines are widely applied in geophysical tasks such as lithofacies analysis and interpretation. However, seismic data processing is often time-consuming and costly due to large input volumes, typically requiring high-performance computing (HPC) clusters. Converting the code for such systems is nontrivial. In this work, we introduce the Accelerated and Scalable Framework (DASF), which accelerates ML pipelines for seismic data processing by leveraging GPU accelerators. DASF supports environments from local machines to HPC clusters with minimal code changes and handles data sets that exceed memory limits. We demonstrated its performance through real-world use cases and found how it improves workflows and enhances seismic data processing.
Faracco et al. (Thu,) studied this question.