用多物理场预训练提升神经算子通用性，实现跨方程迁移。
Towards Universal Neural Operators through Multiphysics Pretraining
- 基于Transformer的神经算子在多物理场数据上预训练
- 跨方程、跨参数场景下仍保持良好泛化能力
- 适合需要快速适配新物理问题的研究者
尽管神经算子广泛用于数据驱动的物理模拟，其训练仍成本高昂。近期进展通过下游学习缓解此问题：先在简单问题上预训练，再在复杂任务上微调。本文研究了此前仅用于特定问题的Transformer神经算子，在更通用的迁移学习框架下的表现。我们在多种PDE问题上评估模型性能，涵盖未见参数的外推、新变量引入以及从多方程数据集迁移。结果表明，先进神经算子架构能有效在不同PDE问题间传递知识。
原文摘要 · Abstract (English)
Although neural operators are widely used in data-driven physical simulations, their training remains computationally expensive. Recent advances address this issue via downstream learning, where a model pretrained on simpler problems is fine-tuned on more complex ones. In this research, we investigate transformer-based neural operators, which have previously been applied only to specific problems, in a more general transfer learning setting. We evaluate their performance across diverse PDE problems, including extrapolation to unseen parameters, incorporation of new variables, and transfer from multi-equation datasets. Our results demonstrate that advanced neural operator architectures can effectively transfer knowledge across PDE problems.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。