Spectral Invariant Learning for Dynamic Graphs under Distribution Shifts (Appendix)

For brevity, we denote the SBM model as SBM(pin,pout), where pin [0,1]C 1 and pout denotes the link probability between the nodes belonging to the same class and the link probability between the nodes from different classes respectively. We adopt C = 5 classes. Based on the class label, each node has two types of parameters flow {0.02,0.04,0.08,0.10,0.12}and