Advances in Machine Learning (ML) are transforming our ability to tackle challenging multiscale subsurface problems. Traditional process-based approaches for characterizing and simulating fluid flow in porous media are time-intensive, labor-intensive, and computationally expensive. Experiments often require long and complex workflows for analyzing data and recognizing complex patterns. Now, ML techniques provide a new paradigm for addressing many classical challenges in subsurface studies from core scale to field scale. This session aims to bring together innovative work from the broad geoscience community. We seek submissions including but not limited to the following topics: (1) supervised learning or science-informed methods for fluid flow simulation and characterization in multi-scale permeable media, (2) real-time scientific inference and data analytics, (3) rapid synthesis of porous media using ML, (4) ML applications in model-based optimization in geosystems management, (5) ML-based uncertainty quantification, history matching, and data assimilation, and (6) continuous updating, meta-learning, and reinforcement learning applications.