Beyond Simulation: Retain-and-Repair Neural Operators for Real-World Adaptation

License: See LICENSE file · Updated 2026-09-30
My rating

This paper introduces the Retain-and-Repair Neural Operator framework for adapting operators pretrained on numerical simulations to real-world data. The pretrained operator is fine-tuned on real data and then frozen to provide source predictions. A shared repair module subsequently learns a sequence of refinements while retaining useful pretrained structure.

Tags: