FAD: Foundation-Model Aided Deep Learning for Marine Seismic Interference Attenuation
Foundation-model-aided deep learning for marine seismic interference attenuation, recorded as accepted by Geophysics on the official PKU faculty profile.
刘马
GeoBrain · People
Jianwei Ma
Team Leader
Boya Distinguished Professor at Peking University and Director of the Center for Artificial Intelligence Geoscience, working across exploration geophysics, applied mathematics and AI.
School of Earth and Space Sciences / Institute for Artificial Intelligence, Peking University
Exploration geophysicsInverse problemsDeep learning
Verified work
Verified publications
Feature-Space Planes Searcher: A Universal Domain Adaptation Framework for Interpretability and Computational Efficiency
A domain-adaptation framework that freezes feature encoders and searches decision boundaries for efficiency and interpretability.
成葛马
GeoFormer: Geometry-Aware Transformer and its application to 5D First-Arrival Picking
A geometry-aware transformer that injects acquisition geometry into prestack seismic representations for robust 5D first-arrival picking.
高马
Personalized Federated Learning via Gaussian Generative Modeling
Gaussian generative modeling of client heterogeneity with dual-scale fusion for personalized federated learning.
胡马
Robust Losses from Univariate Base Functions for Noisy-Label Learning
A systematic framework for constructing analyzable multiclass robust losses from univariate base functions under label noise.
胡马
Spectral Convolutional Neural Networks for Ground Roll Attenuation
A spectral convolutional neural-network approach to coherent ground-roll attenuation, currently recorded as revised.
高马
Towards Foundation Model-Driven Geophysical Inversion by Low-Rank Adaptation and Attention Gates
A parameter-efficient velocity-model-building workflow combining attention gates and enhanced LoRA to adapt multimodal initial velocity, RTM and sparse well information.
马
Foundation Models for Exploration Geophysics
A review of the hierarchy, development workflow, applications and challenges of foundation models for exploration geophysics, including agents and copilots.
刘马
Interpolation-Inspired Self-Supervised Method for Seismic Deblending
An interpolation-inspired self-supervised formulation for seismic deblending that learns interference-attenuation representations without complete labels.
胡马
Seismology Modeling Agent: A Smart Assistant for Geophysical Researchers
SPECFEM MCP serversMCP server suites expose SPECFEM 2D, 3D Cartesian and 3D Globe parameter, meshing, solver and visualization workflows to automated and human-in-the-loop agents.
马
Generative Interpolation via Diffusion Probabilistic Model
A diffusion-probabilistic approach that learns seismic data distributions for generative missing-trace interpolation.
刘马
HC-MVSNet: A Probability Sampling-Based Multi-View-Stereo Network with Hybrid Cascade Structure for 3D Reconstruction
A hybrid-cascade multi-view stereo network that decouples feature resolution from depth intervals and uses probability sampling for accurate, complete 3D reconstruction.
高马
FWIGAN: Full-Waveform Inversion via a Physics-Informed Generative Adversarial Network
FWIGAN implementationA physics-informed adversarial formulation that updates velocity models through an acoustic-wave-equation generator.
马
Sensing Prior Constraints in Deep Neural Networks for Solving Exploration Geophysical Problems
Three strategies for embedding geological and geophysical prior constraints into data, neural layers and learning objectives.
马
Deep Learning for Geophysics: Current and Future Trends
A broad review of deep learning across exploration geophysics, seismology and remote sensing, with challenges and future directions.
马
Deep Learning for Denoising
MathGeo toolboxA systematic study of supervised deep learning for seismic denoising and the role of data and label quality.
马
Deep-Learning Inversion: A Next-Generation Seismic Velocity Model Building Method
FCNVMB implementationA fully convolutional mapping from multishot seismic records to velocity models for rapid next-generation model building.
马
Seismic Data Restoration via Data-Driven Tight Frame
MathGeo toolboxAdaptive tight-frame filters learned from the current estimate for sparse restoration of missing seismic traces.
马
Seismic Data Reconstruction via Matrix Completion
A low-rank matrix-completion formulation for reconstructing sparsely and incompletely sampled seismic traces.
马
The Curvelet Transform
A review of the multiscale, multidirectional curvelet transform and its applications in imaging and scientific computing.
马