Meta networks also follow a similar training procedure to that of matching networks. Here training data is divided into two types: the support set, S = and the test set, U = .
Remember that at present, we have four networks, (), and four sets of model parameters to learn, (θ,ϕ,w,v). Let's go through all of the steps of the algorithm:
Following are the steps of the algorithm:
- Sample K random pairs from support the set, S.
For :