Answers are grounded in this paper and cite their sources.
This paper establishes the optimal approximation error characterization of deep rectified linear unit (ReLU) networks for smooth functions in terms of both width and depth simultaneously. To that end, we first prove that multivariate polynomials can be approximated by deep ReLU networks of width O(N) and depth O(L) with an approximation error O(N-L). Through local Taylor expansions and their deep ReLU network approximations, we show that deep ReLU networks of width O(Nln N) and depth O(Lln L) can approximate f∈ Cˢ([0,1]ᵈ) with a nearly optimal approximation error O(\|f\|Cˢ([0,1]ᵈ)N-2s/dL-2s/d). Our estimate is nonasymptotic in the sense that it is valid for arbitrary width and depth specified by N⁺ and L⁺, respectively.
No takes yet. Share an insight, caveat, or question.
Lu et al. (2021) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: