Encoding NN as MILP Affine layer: direct. y=Wx+b ReLU layer: Pre-condition: Normally box constraints like xi−ϵ≤xi≤xi+ϵ Post-condition:
