Efficient gradient computation of the Jacobian determinant term is a core
problem in many machine learning settings, and especially so in the normalizing
flow framework. Most proposed flow models therefore either
本文介绍了归一化流在高维矩阵行列式计算和神经网络可逆变换这两个应用挑战中的巧妙应用，并使用三孔梁桥、拱桥、斜拉桥和悬索桥的对称结构图像数据集构建和训练了基于 TensorFlow Probability 库中的 Glow API 的归一化流模型，使其能够将桥梁数据集的复杂分布平滑地转换为标准正态分布，并通过对获得的潜在空间进行采样，生成与训练数据集不同的新桥梁类型。
Flow-based deep generative models can be used for novelty detection in time series data and outperform traditional methods like the Local Outlier Factor.