中文

多任务与多算子学习的泛化界限与统计保证:基于 MNO 网络

机器学习 2026-04-03 v1

摘要

多算子学习关注学习算子族 {G[α]:UV}αW\{G[\alpha]:U\to V\}_{\alpha\in W},其中由算子描述符 α\alpha 索引。训练数据按层次结构收集:首先抽样算子实例 α\alpha,随后抽样每个实例的输入函数 uu,最后抽样每个输入的评估点 xx,产生对 G[α][u](x)G[\alpha][u](x) 的带噪声观测。尽管近期工作发展出表达性强的多任务和多算子学习架构及近似理论规模定律,但量化统计泛化保证仍有限。我们提供基于覆盖数的数字泛化分析,用于可分离模型,聚焦 Multiple Neural Operator (MNO) 架构:我们首先推导给定由深度 ReLU 子网络乘积组成的线性组合构成的假设类的显式度量熵界限,然后将这些复杂度界限与 MNO 的近似保证相结合,获得 MNO 在新见的三元组 (α,u,x)(\alpha,u,x) 上期望测试误差的显式近似-估计权衡。 resulting bound makes the dependence on the hierarchical sampling budgets (nα,nu,nx)(n_\alpha,n_u,n_x) transparent and yields an explicit learning-rate statement in the operator-sampling budget nαn_\alpha, providing a sample-complexity characterization for generalization across operator instances. The structure and architecture can also be viewed as a general purpose solver or an example of a "small'' PDE foundation model, where the triples are one form of multi-modality.

关键词

引用

@article{arxiv.2604.01961,
  title  = {Generalization Bounds and Statistical Guarantees for Multi-Task and Multiple Operator Learning with MNO Networks},
  author = {Adrien Weihs and Hayden Schaeffer},
  journal= {arXiv preprint arXiv:2604.01961},
  year   = {2026}
}