How Implicit Regularization of ReLU Neural Networks Characterizes the Learned Function -- Part I: the 1-D Case of Two Layers with Random First Layer
arXiv:1911.02903
Abstract
In this paper, we consider one dimensional (shallow) ReLU neural networks in which weights are chosen randomly and only the terminal layer is trained. First, we mathematically show that for such networks L2-regularized regression corresponds in function space to regularizing the estimate's second derivative for fairly general loss functionals. For least squares regression, we show that the trained network converges to the smooth spline interpolation of the training data as the number of hidden nodes tends to infinity. Moreover, we derive a novel correspondence between the early stopped gradient descent (without any explicit regularization of the weights) and the smoothing spline regression.
arXiv commentadding Appendix C for more intuition, fixing typos, improving formulations, (moving end of Section 3.1 into Appendix B)
References in corpus (12)
- Adam: A Method for Stochastic Optimization
- Neural Tangent Kernel: Convergence and Generalization in Neural Networks
- A Mean Field View of the Landscape of Two-Layers Neural Networks
- Provable approximation properties for deep neural networks
- Implicit Regularization in Deep Learning
- Gradient Dynamics of Shallow Univariate ReLU Networks
- Gradient Descent Quantizes ReLU Network Features
- Implicit Regularization of Discrete Gradient Dynamics in Linear Neural Networks
- A Function Space View of Bounded Norm Infinite Width ReLU Nets: The Multivariate Case
- How do infinite width bounded norm networks look in function space?
- Theory IIIb: Generalization in Deep Networks
- Implicit Regularization in Over-parameterized Neural Networks