We present the SGL learning framework to learn stable graph structures from heterogeneous confounded environments.
