Search

Deep learning has made profound impacts in the domains of data mining and AI, distinguished by the groundbreaking achievements in numerous real-world applications and the innovative algorithm design philosophy. However, it suffers from the inconsistency issue between optimization and generalization, as achieving good generalization, guided by the bias-variance trade-off principle, favors under-parameterized networks, whereas ensuring effective convergence of gradient-based algorithms demands over-parameterized networks. To address this issue, we develop a novel sketching scheme based on deep net components for various tasks. Specifically, we use deep net components with specific efficacy to build a sketching basis that embodies the advantages of deep networks. Subsequently, we transform deep net training into a linear empirical risk minimization problem based on the constructed basis, successfully avoiding the complicated convergence analysis of iterative algorithms. The efficacy of the proposed component-based sketching is validated through both theoretical analysis and numerical experiments. Theoretically, we show that the proposed component-based sketching provides almost optimal rates in approximating saturated functions for shallow nets and also achieves almost optimal generalization error bounds. Numerically, we demonstrate that, compared with the existing gradient-based training methods, component-based sketching possesses superior generalization performance with reduced training costs.

We investigate the expressive power of depth-2 bandlimited random neural networks. A random net is a neural network where the hidden layer parameters are frozen with random assignment, and only the output layer parameters are trained by loss minimization. Using random weights for a hidden layer is an effective method to avoid non-convex optimization in standard gradient descent learning. It has also been adopted in recent deep learning theories. Despite the well-known fact that a neural network is a universal approximator, in this study, we mathematically show that when hidden parameters are distributed in a bounded domain, the network may not achieve zero approximation error. In particular, we derive a new nontrivial approximation error lower bound. The proof utilizes the technique of ridgelet analysis, a harmonic analysis method designed for neural networks. This method is inspired by fundamental principles in classical signal processing, specifically the idea that signals with limited bandwidth may not always be able to perfectly recreate the original signal. We corroborate our theoretical results with various simulation studies, and generally, two main take-home messages are offered: (i) Not any distribution for selecting random weights is feasible to build a universal approximator; (ii) A suitable assignment of random weights exists but to some degree is associated with the complexity of the target function., Comment: Published as a conference paper at ICML 2023