arXiv：cs.LG· Callum Lau, Jeremias Knoblauch, Louis Sharrock·· 3 小时前AI 评分27
PrO-GPs：面向预测的高斯过程后验
Predictively Oriented Gaussian Process Posteriors
AI 导读
研究者提出 Predictively Oriented Gaussian Processes（PrO-GPs），将预测不确定性作为首要推断目标，为存在模型误设的标准高斯过程提供更稳健的替代方案。由于非参数模型的 PrO 后验无法直接计算，作者推导出简化形式与实用采样方案以实现高效计算。合成与真实数据实验显示，在模型误设下 PrO-GPs 的预测分布校准优于标准 GP 方法。
正文
Abstract:Gaussian Processes (GPs) are a powerful tool for modelling and quantifying uncertainty in functional relationships. However, they require practitioners to make a number of design decisions, such as the choice of the kernel and the observation model. Suboptimal choices can produce misspecified models that do not capture the underlying data generating process. We introduce Predictively Oriented Gaussian Processes (PrO-GPs), which treat predictive uncertainty as the primary inferential target and provide a robust alternative to standard GPs. Although direct computation of a PrO posterior for nonparametric models is intractable, we derive a reduced formulation and practical sampling scheme for efficient computation. Through synthetic and real data experiments, we show that PrO-GPs produce better calibrated predictive distributions under model misspecification compared to standard GP approaches.
| Subjects: | Machine Learning (stat.ML); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.03201 [stat.ML] |
| (or arXiv:2610.03201v1 [stat.ML] for this version) | |
| https://doi.org/10.48550/arXiv.2610.03201 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Callum Lau [view email]
[v1]
Fri, 2 Oct 2026 12:16:27 UTC (16,354 KB)
来源：arXiv：cs.LG · arxiv.org