1 problem

Let normalized deep self-attention networks be parametrized via $(\mathbf{M},L)$, where these parameters encode the network as defined in the paper. The normalization removes layer…