neural networks are increasingly being used to solve partial differential
equations (PDEs), replacing slower numerical solvers. However, a critical issue
is that neural PDE solvers require high-quality ground truth data, which
usually must come from the very solvers they are designed t
本文介绍了一个通用的对称化协议，用于向任何给定模型添加旋转等变性，同时保留所有其他约束条件。我们将此方法应用于 Point Edge Transformer（PET）体系结构，并展示其在多个分子和固体基准数据集上实现了最先进的性能。通过减轻模型内显式纳入旋转对称性的需要，我们的方法弥合了不同社区所使用的方法之间的差距，并简化了化学和材料建模的深度学习方案的设计。