因果表征學習

因果圖 (Causal Diagram)

如果整個 DAG 的結構已知且所有的變量都可觀測，那么我們可以根據上面 do 算子的公式算出任意變量之間的因果作用。但是，在絕大多數的實際問題中，我們既不知道整個 DAG 的結構，也不能將所有的變量觀測到。

“后門準則”（backdoor criterion）和“前門準則”（frontdoor criterion）

后門準則：Z可以同時影響or產生 X/Y，那么Z就相當于X/Y因果關系的后門（不影響X-> Y之間，影響兩者），Z也是X/Y的混雜因子
前門準則：Z影響X-> Y的前門路徑（直接影響X->Y）

后門調整

Z在X-> Y的后門路徑上，那么一般會利用do算子進行干預，進行【后門調整】，可以看到這里從do(X) -> (X,z)

Visual Commonsense R-CNN

Causal Intervention for Weakly-Supervised Semantic Segmentation

Causal Intervention and Parameter-Free Reasoning for Few-Shot SAR Target Recognition

A Domain Generalization Network Exploiting Causal Representations and Non-Causal Representations for Three-Phase Converter Fault Diagnosis

Causal Prototype-Inspired Contrast Adaptation for Unsupervised Domain Adaptive Semantic Segmentation of High-Resolution Remote Sensing Imagery

Causality-Inspired Single-Source Domain Generalization for Medical Image Segmentation

Generalizable Single-Source Cross-Modality Medical Image Segmentation via Invariant Causal Mechanisms

Causality-inspired Unsupervised Domain Adaptation with Target Style Imitation for Medical Image Segmentation

CPI-Parser: Integrating Causal Properties Into Multiple Human Parsing

Revisiting Few-Shot Learning From a Causal Perspective

Causal Meta-Transfer Learning for Cross-Domain Few-Shot Hyperspectral Image Classification