Guardado en:
Detalles Bibliográficos
Autores principales: Queiruga, Alejandro Francisco, Gutman-Solo, Theo, Jiang, Shuai
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2506.15199
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917532528541696
author Queiruga, Alejandro Francisco
Gutman-Solo, Theo
Jiang, Shuai
author_facet Queiruga, Alejandro Francisco
Gutman-Solo, Theo
Jiang, Shuai
contents While there are many applications of ML to scientific problems that look promising, visuals can be deceiving. Using numerical analysis techniques, we rigorously quantify the accuracy, convergence rates, and generalization bounds of certain ML models applied to linear differential equations for parameter discovery or solution finding. Beyond the quantity and discretization of data, we identify that the function space of the data is critical to the generalization of the model. A similar lack of generalization is empirically demonstrated for commonly used models, including physics-specific techniques. Counterintuitively, we find that different classes of models can exhibit opposing generalization behaviors. Based on our theoretical analysis, we also introduce a new mechanistic interpretability lens on scientific models whereby Green's function representations can be extracted from the weights of black-box models. Our results inform a new cross-validation technique for measuring generalization in physical systems, which can serve as a benchmark.
format Preprint
id arxiv_https___arxiv_org_abs_2506_15199
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Interpretability and Generalization Bounds for Learning Spatial Physics
Queiruga, Alejandro Francisco
Gutman-Solo, Theo
Jiang, Shuai
Machine Learning
While there are many applications of ML to scientific problems that look promising, visuals can be deceiving. Using numerical analysis techniques, we rigorously quantify the accuracy, convergence rates, and generalization bounds of certain ML models applied to linear differential equations for parameter discovery or solution finding. Beyond the quantity and discretization of data, we identify that the function space of the data is critical to the generalization of the model. A similar lack of generalization is empirically demonstrated for commonly used models, including physics-specific techniques. Counterintuitively, we find that different classes of models can exhibit opposing generalization behaviors. Based on our theoretical analysis, we also introduce a new mechanistic interpretability lens on scientific models whereby Green's function representations can be extracted from the weights of black-box models. Our results inform a new cross-validation technique for measuring generalization in physical systems, which can serve as a benchmark.
title Interpretability and Generalization Bounds for Learning Spatial Physics
topic Machine Learning
url https://arxiv.org/abs/2506.15199