Radial basis operator networks
Authors
Loading...
Publisher link
Rights
This is an accepted version of a paper presented at the 7th International Conference on Advances in Signal Processing and Artificial Intelligence (ASPAI' 2025). © 2025 IFSA Publishing. Used with permission.
Abstract
Operator networks are designed to approximate nonlinear operators, which map between infinite-dimensional spaces like function spaces. These networks are increasingly important in machine learning, particularly in scientific computing, due to their ability to handle data common in fields like climate modeling and fluid dynamics, where inputs are often discretized continuous fields (e.g., temperature or velocity distributions). We introduce the radial basis operator network (RBON), which represents a breakthrough as the first operator network capable of learning an operator in both the time and frequency domains when adjusted to accept complex-valued inputs. Despite the small, single hidden-layer structure, the RBON boasts small ܮଶ relative test error for both in- and out-of-distribution data (OOD) of less than 1 × 10ି in some benchmark cases. Furthermore, it maintains small errors on OOD data from entirely different function classes than those used during training, showcasing its robustness and adaptability for advanced scientific applications.
Citation
Kurz, J., Oughton, S., & Liu, S. (2025). Radial basis operator networks. In S. Y. Yurish (Ed.), Proceedings of the 7th International Conference on Advances in Signal Processing and Artificial Intelligence (ASPAI' 2025), Innsbruck, Austria, 8-10 April 2025 (pp. 189-196). IFSA Publishing. https://doi.org/10.13140/RG.2.2.14725.46566
Series name
Date
Publisher
IFSA Publishing
Source Title
Proceedings of the 7th International Conference on Advances in Signal Processing and Artificial Intelligence (ASPAI' 2025)