STAMO adopts a two-stages training strategy. In stage 1, pretraining graph attention network to produce coarse aligned embeddings. In stage 2, identifying anchors via Fused Gromov-Wasserstein optimal ...