{"id":"arxiv/2606.18305","url":"https://arxiv.org/abs/2606.18305","tags":["math.NA","cs.LG"],"score":0,"title":"Starter-Iterator Neural Operator: A Unified Architecture for High-Fidelity Forward and Inverse PDE Problems","domain":"math.NA","source":"arxiv","authors":["Kuilin Qin","Lianfang Wang","Xu Sun","Jiwei Jia","Yu Wang","Yong Wang","Yuping Duan"],"reviews":0,"subject":"math.NA","summary":"Operator learning is an emerging field at the intersection of machine learning and scientific computing. By learning mappings between function spaces, neural operators provide data-driven surrogate models for families of partial differential equations (PDEs). Once trained, these models can evaluate solution operators efficiently, making them suitable for many-query applications such as real-time prediction and parameter sweeps. However, maintaining high approximation accuracy and stable long-term predictions remains challenging for complex forward and inverse problems. To address these…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-16","reactions":{}}