arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2301.08893cs.LGphysics.app-ph
收藏

具有E(n)-等方差的空间注意动力网络

Spatial Attention Kinetic Networks with E(n)-Equivariance

Weill Cornell Medical College(威尔康奈尔医学院);Sloan Kettering Institute(斯隆凯特琳研究所);Memorial Sloan Kettering Cancer Center(纪念斯隆凯特琳癌症中心)

Yuanqing Wang, John D. Chodera

摘要:

神经网络在n维几何空间上的旋转、平移、反射和置换等变,在物理建模中显示出了希望,例如精确但廉价地建模复杂的势能表面,以指导复杂动力系统的采样或预测它们的时间演化。目前最先进的方法采用球谐函数来编码粒子之间的高阶相互作用,这在计算上是昂贵的。在本文中,我们提出了一个简单的替代功能形式,使用神经参数化的线性组合的边缘向量,以实现等方差,同时仍然普遍近似节点环境。为了证明这一点,我们设计了具有E(n)-等方差(SAKE)的空间注意力动力学网络,它在多体系统建模任务中具有竞争力,同时速度明显更快。

英文摘要:

Neural networks that are equivariant to rotations, translations, reflections, and permutations on n-dimensional geometric space have shown promise in physical modeling for tasks such as accurately but inexpensively modeling complex potential energy surfaces to guiding the sampling of complex dynamical systems or forecasting their time evolution. Current state-of-the-art methods employ spherical harmonics to encode higher-order interactions among particles, which are computationally expensive. In this paper, we propose a simple alternative functional form that uses neurally parametrized linear combinations of edge vectors to achieve equivariance while still universally approximating node environments. Incorporating this insight, we design spatial attention kinetic networks with E(n)-equivariance, or SAKE, which are competitive in many-body system modeling tasks while being significantly faster.

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。

↑