In the following experiments, unless otherwise explicitly stated,we use the DMGC model asthe graphmatchingmethodinthissection. (a) Metastepsizeβ (b) # SamplesN (c) Initialbandwidthb0 (d) Parameters

We also include equal number of randomly selected disconnected links that servers as negative samples. The performance curves initially raise and then drop quickly whenβ continuously increases. This demonstrates that there must exist the optimalβ that makes the meta learning be maximally 20 optimized. Sensitivity of number of samples N. Figure 7 (b) exhibits the sensitivity ofN in our MLPGD model withN between 1 and 15. All models were trained for 500 iterations, with a batch size of 512, and a learningrateof0.001.