ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2008.07648
22
1

Nonparametric Learning of Two-Layer ReLU Residual Units

17 August 2020
Zhunxuan Wang
Linyun He
Chunchuan Lyu
Shay B. Cohen
    MLT
    OffRL
ArXivPDFHTML
Abstract

We describe an algorithm that learns two-layer residual units using rectified linear unit (ReLU) activation: suppose the input x\mathbf{x}x is from a distribution with support space Rd\mathbb{R}^dRd and the ground-truth generative model is a residual unit of this type, given by y=B∗[(A∗x)++x]\mathbf{y} = \boldsymbol{B}^\ast\left[\left(\boldsymbol{A}^\ast\mathbf{x}\right)^+ + \mathbf{x}\right]y=B∗[(A∗x)++x], where ground-truth network parameters A∗∈Rd×d\boldsymbol{A}^\ast \in \mathbb{R}^{d\times d}A∗∈Rd×d represent a full-rank matrix with nonnegative entries and B∗∈Rm×d\boldsymbol{B}^\ast \in \mathbb{R}^{m\times d}B∗∈Rm×d is full-rank with m≥dm \geq dm≥d and for c∈Rd\boldsymbol{c} \in \mathbb{R}^dc∈Rd, [c+]i=max⁡{0,ci}[\boldsymbol{c}^{+}]_i = \max\{0, c_i\}[c+]i​=max{0,ci​}. We design layer-wise objectives as functionals whose analytic minimizers express the exact ground-truth network in terms of its parameters and nonlinearities. Following this objective landscape, learning residual units from finite samples can be formulated using convex optimization of a nonparametric function: for each layer, we first formulate the corresponding empirical risk minimization (ERM) as a positive semi-definite quadratic program (QP), then we show the solution space of the QP can be equivalently determined by a set of linear inequalities, which can then be efficiently solved by linear programming (LP). We further prove the strong statistical consistency of our algorithm, and demonstrate its robustness and sample efficiency through experimental results on synthetic data and a set of benchmark regression datasets.

View on arXiv
Comments on this paper