Search papers, labs, and topics across Lattice.
This paper investigates Gaussian regression using the Parhi-Nowak deep-RBV^2 architecture, revealing that the minimax risk has a quadratic dependence on depth, which is intrinsic to the architecture under specific sample-size conditions. The authors establish a local packing argument that demonstrates the relationship between depth and risk, supported by a bias-corrected approximation theorem and balanced amplification techniques. Key findings indicate that the architecture's performance is significantly influenced by depth, with implications for understanding the limits of representation in deep learning models.
The minimax risk in deep Gaussian regression exhibits a surprising quadratic dependence on depth, challenging conventional assumptions about model capacity.
We study Gaussian regression over the explicit vector-valued Parhi--Nowak deep-RBV^2 architecture with depth L, width w, layer-sum variation budget A, and output bound B. For this O(L w^2)-parameterized architecture, the known lower and upper bounds differ by one factor of depth. We construct a local packing showing that the quadratic depth dependence is intrinsic under an explicit sample-size-dependent radius condition. The packing has log-cardinality Omega(L^2 w^2 log w); its codewords lie in an O(lambda) L^2 ball and are pairwise Omega(lambda)-separated. The main ingredients are a bias-corrected bounded-coefficient approximation theorem and balanced amplification: multiplying a depth-D ReLU network by q can be implemented using one constant channel so that every coefficient grows by only q^(1/D). Translation to vector-valued RBV^2 blocks then has layer-sum cost O(D w^2 q^(1/D)). Gaussian Fano yields a radius-explicit lower bound governed by the output, testing, and representation scales. Under A=B=R, sigma proportional to R, and the stated radius condition, this gives minimax risk at least of order L^2 w^2 log(w) R^2/n. A pseudodimension-based finite-net upper bound gives O-tilde(L^2 w^2 R^2/n) for unbounded Gaussian responses. Thus the minimax risk has quadratic polynomial dependence on depth, up to logarithmic factors, and exhibits a transition to representation-limited behavior at smaller radius.