Advances in Data-Intensive and Computational Astrophysics: Machine Learning, HPC, and Statistical Inference

Abstract:

The past decade has witnessed an exponential growth in astronomical data volume—driven by facilities like the Legacy Survey of Space and Time (LSST), LIGO-Virgo-KAGRA (LVK), and Euclid—creating a “data revolution” that demands advanced computational tools. This review synthesizes 2022–2025 progress in data-intensive and computational astrophysics, focusing on four core areas: (1) big data analytics with machine learning (ML), including transformer-based models for LSST supernova classification (accuracy >98%); (2) AI-assisted image processing, such as deep learning for Hubble Space Telescope (HST) artifact removal (signal-to-noise improvement >40%); (3) high-performance computing (HPC) for large-scale simulations, e.g., exascale cosmic structure models with 100 billion particles; (4) statistical inference techniques, including Bayesian neural networks for gravitational wave (GW) parameter estimation (uncertainty reduction ~30%). We present a “Multi-Task Computational Framework” that integrates these tools, validated by applications to 10+ astronomical datasets (e.g., LSST galaxy catalogs, LVK GW events). We also discuss challenges like data heterogeneity and computational scalability, and outline future priorities—including quantum machine learning for real-time data processing and edge computing for space-based observatories—to address the next generation of astronomical data challenges.