CauseFormer: An Interpretable Transformer for Temporal Causal Discovery
Published in Under Review, 2023
Temporal causal discovery has become an effective technique to reveal the internal causality of time series. However, most existing deep learning-based causal discovery methods only capture causal relations by analyzing the parameters of some components in the model, e.g., attention weights and convolution weights, which is a local-level mapping process from the parameters to the causality and fails to investigate the structure of the whole model to discover the causality. To facilitate the glob…
Recommended citation: Under Review