MSPT：通过并行化多尺度注意力实现高效大规模物理建模
MSPT: Efficient Large-Scale Physical Modeling via Parallelized Multi-Scale Attention
把「图像块内局部点注意力 + 粗粒度图像块级全局注意力」的双尺度注意力，配合球树做不规则几何的空间划分，可直接迁移到百万级点云或高斯球的场景表示与神经场建模中，在单卡上显著降低显存占用。
原文依据 combines local point attention within patches with global attention to coarse patch-level representations
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论面向工业级物理仿真的神经求解器面临一个关键的可扩展性挑战：如何高效地同时捕捉数百万空间单元上的细粒度局部交互与长程全局依赖。我们提出多尺度图像块 Transformer（MSPT），该架构将图像块内的局部点注意力与针对粗粒度图像块级表示的全局注意力相结合。为将输入域划分为空间连贯的图像块，我们采用球树，从而高效处理不规则几何。这种双尺度设计使 MSPT 能够在单块 GPU 上扩展到数百万个点。我们在标准 PDE 基准（弹性、塑性、流体动力学、多孔流）以及大规模空气动力学数据集（ShapeNet-Car、Ahmed-ML）上验证了 MSPT，在显著降低显存占用与计算成本的同时取得了最优精度。
英文原文摘要 882 字符
A key scalability challenge in neural solvers for industrial-scale physics simulations is efficiently capturing both fine-grained local interactions and long-range global dependencies across millions of spatial elements. We introduce the Multi-Scale Patch Transformer (MSPT), an architecture that combines local point attention within patches with global attention to coarse patch-level representations. To partition the input domain into spatially-coherent patches, we employ ball trees, which handle irregular geometries efficiently. This dual-scale design enables MSPT to scale to millions of points on a single GPU. We validate MSPT on standard PDE benchmarks (elasticity, plasticity, fluid dynamics, porous flow) and large-scale aerodynamic datasets (ShapeNet-Car, Ahmed-ML), achieving state-of-the-art accuracy with substantially lower memory footprint and computational cost.