About identifiability of Notears MLP

[MingjiaLi666]

Hello Mr. Zheng,
recently I' ve read your paper Notears MLP, excellent work!
However as a consensus in causal discovery community, the ANM model is identifiable since we can test whether the regression residual is independent of input.
The weird thing is that your Notears MLP done very well on ANM with mlp nonlinear, but it seems there is no operation about the residual, so how could it achieve identifiability.
Could you please answer this question, thx a lot for your time!