. Unlike existing score-based diffusion models, the reverse process can take observations (on the top left of

the fifigure) as a conditional input, allowing the model to exploit information in the observations for

denoising. We utilize an attention mechanism to capture the temporal and feature dependencies of

time series

本文利用一种注意机制来捕捉时间序列的时间依赖性和特征依赖性

在实践中，我们并不知道ground truth missing value（i.e., imputation targets），或训练数据可能根本不包含缺失值。然后，受masked language modeling的启发，本文开发了一种自监督（self-supervised）的训练方法，将观察值分离成条件信息和计算目标