论文标题
Wasserstein-2内核的等级高斯流程
Hierarchical Gaussian Processes with Wasserstein-2 Kernels
论文作者
论文摘要
堆叠高斯工艺会严重降低模型检测异常值的能力，当与非零平均功能结合使用时，该模型将其进一步推断出低的非参数方差到低训练数据密度区域。我们提出了一种在欧几里得和瓦斯坦斯坦空间中运行的Varifold理论启发的混合核。我们认为，直接考虑到Wasserstein-2距离计算的差异对于在整个层次结构中保持异常地位至关重要。我们在中型和大型数据集上显示出改进的性能，并增强了玩具和真实数据的分布外检测。
Stacking Gaussian Processes severely diminishes the model's ability to detect outliers, which when combined with non-zero mean functions, further extrapolates low non-parametric variance to low training data density regions. We propose a hybrid kernel inspired from Varifold theory, operating in both Euclidean and Wasserstein space. We posit that directly taking into account the variance in the computation of Wasserstein-2 distances is of key importance towards maintaining outlier status throughout the hierarchy. We show improved performance on medium and large scale datasets and enhanced out-of-distribution detection on both toy and real data.
