Correcting Neural Operator Spectral Bias via Diffusion Posterior Sampling with Sparse Observations
作者: Niccolò Perrone, Fanny Lehmann, Stefania Fresca, Filippo Gatti
分类: cs.LG, physics.geo-ph
发布日期: 2026-06-02
🔗 代码/项目: GITHUB
💡 一句话要点
提出FreqNO-DPS以解决神经算子谱偏差问题
🎯 匹配领域: 支柱六:视频提取与匹配 (Video Extraction)
关键词: 神经算子 谱偏差 扩散后验采样 稀疏观测 高频内容 偏微分方程 预测精度
📋 核心要点
- 现有的神经算子代理方法在处理高频内容时存在谱偏差,影响了其在细节丰富场景中的可靠性。
- 论文提出FreqNO-DPS方法,通过扩散后验采样框架,结合稀疏观测和频率依赖的指导分数,克服了谱偏差问题。
- 在3D弹性波场预测实验中,该方法在低传感器覆盖率下实现了接近零的谱偏差,显著提升了预测精度。
📝 摘要(中文)
神经算子代理(NO)能够比数值求解器更快地近似偏微分方程(PDE)解,但存在谱偏差,导致高频内容被系统性衰减,限制了在细尺度结构重要的情况下的可靠性。论文提出了一种新的方法FreqNO-DPS,通过将NO预测视为扩散后验采样框架中的辅助观测,结合无条件的基于分数的扩散先验和稀疏观测条件下的扩散后验采样,解决了这一问题。实验结果表明,该方法在3D弹性波场预测中,能够在5%和2%的传感器覆盖率下,实现几乎零的谱偏差。
🔬 方法详解
问题定义:论文旨在解决神经算子代理(NO)在近似偏微分方程解时的谱偏差问题。现有方法在高频内容处理上存在系统性衰减,限制了其在细节丰富场景中的应用。
核心思路:提出FreqNO-DPS方法,将NO预测视为扩散后验采样中的辅助观测,结合无条件的扩散先验和稀疏观测条件,设计频率依赖的指导分数以克服谱偏差。
技术框架:该方法包括几个主要模块:首先,训练无条件的基于分数的扩散先验;然后,在稀疏观测条件下进行扩散后验采样;最后,通过频率依赖的指导分数来调整NO的输出,确保高频内容的准确性。
关键创新:最重要的创新在于提出了频率依赖的指导分数,这种设计避免了传统方法中需要反向传播去噪器的复杂性,直接通过频率加权来提升预测精度。
关键设计:方法中使用了配对的代理/参考数据,损失函数设计为考虑频率依赖性,网络结构上采用了冻结的神经算子以指导采样过程,确保了在新代理上也能验证其有效性。
🖼️ 关键图片
📊 实验亮点
在3D弹性波场预测实验中,FreqNO-DPS方法在5%和2%的传感器覆盖率下实现了几乎零的谱偏差,显著优于传统的神经算子代理和仅基于传感器的扩散后验采样,验证了频率依赖校准的重要性。
🎯 应用场景
该研究的潜在应用领域包括地震波预测、气候模型和其他需要高精度偏微分方程求解的科学计算。通过减少谱偏差,该方法能够在实际应用中提供更可靠的预测结果,提升模型在复杂场景下的表现,未来可能推动相关领域的研究进展。
📄 摘要(原文)
Neural operator surrogates (NO) approximate PDE solutions orders of magnitude faster than numerical solvers, but suffer from spectral bias: high-frequency content is systematically attenuated, limiting reliability where fine-scale structure matters. Sparse sensor measurements of the field are often available too, offering pointwise accuracy without spectral distortion but covering only a small fraction of the domain. We address this by treating NO predictions as auxiliary observations in a diffusion posterior sampling framework. Our method, FreqNO-DPS (https://github.com/niccoloperrone/FreqNO-DPS), combines an unconditional score-based diffusion prior, trained on high-fidelity simulations, with diffusion posterior sampling (DPS) conditioned on sparse observations and guided by a frozen neural operator. Naive integration reintroduces the surrogate's spectral bias; we resolve this with a closed-form, spectrally shaped guidance score that weights the surrogate by its frequency-dependent accuracy and needs no denoiser backpropagation. A distribution-free analysis bounds the approximation error across the frequency-diffusion-time plane and shows the guidance's frequency dependence is preserved regardless of distributional assumptions. On 3D elastic wavefield prediction at 5% and 2% sensor coverage, the method reaches near-zero spectral bias across all bands, where both the surrogate and sensor-only DPS show systematic high-frequency attenuation. Isotropic guidance, the natural baseline, improves pointwise accuracy but carries the bias into the posterior nearly intact, confirming that frequency-dependent calibration is essential, not merely beneficial. The framework needs only paired surrogate/reference data and exploits no problem-specific structure beyond the residual's approximate spectral diagonality, verifiable for new surrogates via the coherence diagnostic we provide.