Quadratic Depth Dependence in Deep ReLU Regression

Analysis of the Parhi--Nowak deep-RBV^2 architecture reveals that local entropy exhibits a quadratic dependence on depth L. The findings show that for O(L w^2)-parameterized networks, the complexity is constrained by a layer-sum cost of O(D w^2 q^(1/D)) when implementing coefficient amplification.