Algorithm:
Second-order MAML with the zeroing trick
Require: Task distribution
p(T) Require:
α,η: inner loop and meta-learning rates
Require: Randomly initialized base-model parameters
θ
1: Set w ← 0 (the zeroing trick) 2: while not done do 3: Sample tasks {T1,…TNbatch} from p(T) 4: for i = 1,2,…,Nbatch do 5: {Ditr,Ditest}← sample from Ti 6: θi = θ 7: for j = 1,2,…,Nstep do 8: θi ← θi - α∇θiL(θi,Ditr) 9: end for 10: end for 11: Update θ ← θ - η∇θ ∑ i=1NbatchL(θ i,Ditest) 12: Set w ← 0 (the zeroing trick) 13: end while