Iterative Refinement Neural Operators are Learned Fixed-Point Solvers: A Principled Approach to Spectral Bias Mitigation

In three linesIRNO (Iterative Refinement Neural Operator) enhances neural operators with an iterative refinement module using fixed-point iteration theory. A progressive spectral loss explicitly targets high-frequency errors. Results: 56% improvement on turbulent flow, error reduction to 1.48-2.04% in high frequencies on Active Matter.

Summary generated by Claude — human-verified