Martingale Neural Operators learn stochastic marginals via Doob-Meyer factorization

Researchers have developed a new neural operator architecture called Martingale Neural Operator (MNO) designed to handle stochastic partial differential equations (SPDEs). Unlike existing deterministic operators that collapse to a conditional mean, MNO leverages the Doob-Meyer theorem to directly map initial conditions to the conditional mean and covariance of the terminal law. This approach allows for efficient uncertainty quantification by recovering variance and tail structure, outperforming a conditional diffusion baseline in terms of Wasserstein distance and speed. AI

影响 Introduces a novel neural operator architecture for improved handling of stochastic systems, potentially advancing uncertainty quantification in scientific modeling.

排序理由 The cluster contains a new academic paper detailing a novel model architecture. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

• Doob-Meyer theorem
• Martingale Neural Operator
• Neural Operators

• 论文
• 模型发布

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →