PREFACE: MACHINE LEARNING FOR LARGE-SCALE SCIENTIFIC DATA

The explosion of large-scale scientific data from simulations, experiments, and observations across disciplines has led to new challenges in data analysis, modeling, and interpretation.Traditional computational approaches, while grounded in physical principles, often struggle to scale with the complexity and volume of modern scientific datasets.Machine learning (ML), particularly in the form of deep learning, has emerged as a transformative tool for accelerating scientific discovery.It offers a data-driven means to approximate governing equations, extract patterns, and reduce the dimensionality of complex systems, all while integrating seamlessly with domain knowledge.This special issue explores recent advances in applying ML techniques to largescale scientific data, showcasing how they can augment, accelerate, and, in some cases, redefine conventional modeling and simulation paradigms.The five papers in this issue present innovative and diverse contributions.Gurung et al. ( 2025) proposes a hybrid model that combines partial differential equations and deep neural networks to simulate calcium dynamics in neurons, enhancing biological realism, and computational efficiency.Huynh et al. ( 2025) develops a parameter estimation framework for fractured porous media using a direct filter method, demonstrating real-time inference of fracture widths from flow data.Chen et al. ( 2025) introduces a piecewise mesh-free neural network for solving Poisson-Boltzmann equations across protein surfaces, addressing challenges in biological electrostatics.Xie (2025) presents a convolutional auto-encoder-based, reduced-order model for the quasi-geostrophic equations, enabling accurate representation of nonlinear geophysical fluid dynamics with significant dimensionality reduction.Finally, Du et al. (2025) proposes the finite expression method-a reinforcement learning-based symbolic regression approach-for learning interpretable models of epidemiological dynamics from data, balancing predictive power with transparency.Together, these works demonstrate the power and promise of machine learning in unlocking insights from large-scale scientific datasets across domains.