{"slug": "iterative-refinement-neural-operators-are-learned-fixed-point-solvers-a-approach", "title": "Iterative Refinement Neural Operators are Learned Fixed-Point Solvers: A Principled Approach to Spectral Bias Mitigation", "summary": "Researchers have developed the Iterative Refinement Neural Operator (IRNO), a new AI architecture that improves scientific modeling by iteratively correcting errors through fixed-point iteration, directly addressing the spectral bias problem that prevents standard neural operators from resolving high-frequency details. The method, which decomposes predictions into coarse initializations followed by successive residual corrections, achieved up to 56.05% lower error on turbulent flow simulations and reduced high-frequency error ratios to as low as 1.48-2.04% in active matter systems. IRNO's progressive spectral loss training strategy enables stable convergence beyond trained iteration counts, offering a principled approach to building more accurate data-driven surrogates for complex physical systems.", "body_md": "arXiv:2605.24041v1 Announce Type: new\nAbstract: Neural operators serve as fast, data-driven surrogates for scientific modeling but typically rely on a monolithic, single-pass inference procedure that struggles to resolve high-frequency details, a limitation known as spectral bias. We introduce the Iterative Refinement Neural Operator (IRNO), which augments pre-trained operators with a learned refinement module iteratively applied via fixed-point iteration. IRNO decomposes the prediction into a coarse initialization followed by successive residual corrections, paralleling classical numerical solvers. Under local assumptions, we establish contraction of the induced operator, ensuring convergence to a unique fixed point. To explicitly target high-frequency errors, we propose a progressive spectral loss that adaptively increases penalty on high-frequency components over refinement steps during training. Across physical systems, IRNO consistently lowers error, with up to 56.05% improvement on turbulent flow. On Active Matter, spectral analysis reveals that, relative to base operator, the normalized error ratios decrease to 27.72-36.10% in low-, 5.07-6.68% in mid-, and 1.48-2.04% in high-frequencies, remaining stable beyond the trained iteration count. Code is available at https://github.com/xiaotianliu-dartmouth/Iterative_Refinement_Neural_Operator", "url": "https://wpnews.pro/news/iterative-refinement-neural-operators-are-learned-fixed-point-solvers-a-approach", "canonical_source": "https://arxiv.org/abs/2605.24041", "published_at": "2026-05-26 04:00:00+00:00", "updated_at": "2026-05-26 04:07:00.589519+00:00", "lang": "en", "topics": ["machine-learning", "neural-networks", "ai-research"], "entities": ["Iterative Refinement Neural Operator", "IRNO", "arXiv", "xiaotianliu-dartmouth"], "alternates": {"html": "https://wpnews.pro/news/iterative-refinement-neural-operators-are-learned-fixed-point-solvers-a-approach", "markdown": "https://wpnews.pro/news/iterative-refinement-neural-operators-are-learned-fixed-point-solvers-a-approach.md", "text": "https://wpnews.pro/news/iterative-refinement-neural-operators-are-learned-fixed-point-solvers-a-approach.txt", "jsonld": "https://wpnews.pro/news/iterative-refinement-neural-operators-are-learned-fixed-point-solvers-a-approach.jsonld"}}