4.Deep Causal Learning: Representation, Discovery and Inference (2021.11–2022.09)

The research highlights the current challenges and constraints within the realm of causal learning, underscoring the necessity of integration with deep learning techniques. Subsequently, it introduces the latest advancements in three fundamental areas: causal representation learning, deep causal discovery, and deep causal inference. By exploring these cutting-edge methods, the study aims to push the boundaries of understanding and harness the power of causal learning in conjunction with deep learning to tackle complex problems effectively.

To the best of our knowledge, this work is the first comprehensive survey of the combination of causal representation, causal discovery, causal inference and deep learning. It contains classic and latest deep causal discovery and deep causal inference methods, such as NOTEARS, DAG-GNN, Grad-DAG; CFR, DragonNet, CEVAE, GANITE, DeepIV, etc.

We mechanistically unveil the enhancements that deep learning brings to causal inference.

More details can be found in the following paper:

Deep causal learning: representation, discovery and inference

Z Deng, X Zheng, H Tian, DD Zeng

arXiv preprint arXiv:2211.03374