arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2512.01738cs.LG
收藏

MSPT:通过并行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

University of Amsterdam(阿姆斯特丹大学)

Pedro M. P. Curvo, Jan-Willem van de Meent, Maksim Zhdanov

AI总结:

MSPT通过并行多尺度注意力机制,高效处理大规模物理模拟中的局部交互与全局依赖,实现高精度低开销的求解

摘要:

神经求解器在工业级物理模拟中的主要可扩展性挑战是高效捕捉数百万空间元素中的细粒度局部交互和长距离全局依赖。我们引入了多尺度补丁变换器(MSPT),这是一种结合补丁内的局部点注意力和全局注意力的架构,用于粗粒度补丁级表示。为将输入域划分为空间上连贯的补丁,我们采用球树,其能高效处理不规则几何结构。这种双尺度设计使MSPT能够单块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)和大规模空气动力学数据集(ShapeNet-Car,Ahmed-ML)上验证了我们的方法,实现了最先进的准确性,同时具有显著更低的内存足迹和计算成本。

英文摘要:

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 our method 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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑