GeoBrain is an academic research group associated with Peking University and Harbin Institute of Technology, working at the intersection of artificial intelligence, applied mathematics, and exploration geophysics. Its goal is to make geophysical analysis more capable and trustworthy through reproducible, interpretable, and transferable methods. The team studies five connected areas: seismic data processing, geophysical inversion, AI algorithms, foundation models for geophysics, and geophysical agents.

Its work addresses challenges such as noisy or incomplete seismic records, subsurface imaging, velocity-model construction, and full-waveform inversion, while exploring methods including generative learning, self-supervision, physics-informed approaches, and domain adaptation. GeoBrain also develops SeismicBench, an open academic benchmark that organizes seismic-processing methods, datasets, papers, and evaluation results to support reproducible research. In agent research, the group has demonstrated workflows using SPECFEM software tools for model setup, meshing, simulation, and visualization, with human collaboration.

The website offers a 78-page introductory guide for new students, shares publications and project updates, and invites academic collaboration