中文
相关论文

相关论文: Expressive Power of ReLU and Step Networks under F…

200 篇论文

Most existing expressivity theories for neural networks assume exact real arithmetic, whereas practical neural networks are executed under finite-precision floating-point arithmetic with implementation-dependent execution semantics. Recent…

机器学习 · 计算机科学 2026-05-28 Yeachan Park , Geonho Hwang , Wonyeol Lee , Sejun Park

The expressive power of neural networks is important for understanding deep learning. Most existing works consider this problem from the view of the depth of a network. In this paper, we study how width affects the expressiveness of neural…

机器学习 · 计算机科学 2017-11-02 Zhou Lu , Hongming Pu , Feicheng Wang , Zhiqiang Hu , Liwei Wang

Existing works on the expressive power of neural networks typically assume real parameters and exact operations. In this work, we study the expressive power of quantized networks under discrete fixed-point parameters and inexact fixed-point…

机器学习 · 计算机科学 2026-01-21 Yeachan Park , Sejun Park , Geonho Hwang

It is difficult to describe in mathematical terms what a neural network trained on data represents. On the other hand, there is a growing mathematical understanding of what neural networks are in principle capable of representing.…

机器学习 · 计算机科学 2025-06-25 Daan Huybrechs

The success of deep networks has been attributed in part to their expressivity: per parameter, deep networks can approximate a richer class of functions than shallow networks. In ReLU networks, the number of activation patterns is one…

机器学习 · 统计学 2019-10-22 Boris Hanin , David Rolnick

Transformer networks have achieved remarkable empirical success across a wide range of applications, yet their theoretical expressive power remains insufficiently understood. In this paper, we study the expressive capabilities of…

机器学习 · 计算机科学 2026-03-04 Linyan Gu , Lihua Yang , Feng Zhou

We study the expressive power of deep ReLU neural networks for approximating functions in dilated shift-invariant spaces, which are widely used in signal processing, image processing, communications and so on. Approximation error bounds are…

机器学习 · 计算机科学 2023-12-05 Yunfei Yang , Zhen Li , Yang Wang

This article is concerned with the approximation and expressive powers of deep neural networks. This is an active research area currently producing many interesting papers. The results most commonly found in the literature prove that neural…

机器学习 · 计算机科学 2019-05-08 I. Daubechies , R. DeVore , S. Foucart , B. Hanin , G. Petrova

The classical universal approximation (UA) theorem for neural networks establishes mild conditions under which a feedforward neural network can approximate a continuous function $f$ with arbitrary accuracy. A recent result shows that neural…

机器学习 · 计算机科学 2026-01-28 Geonho Hwang , Wonyeol Lee , Yeachan Park , Sejun Park , Feras Saad

Researchers have developed neural network verification algorithms motivated by the need to characterize the robustness of deep neural networks. The verifiers aspire to answer whether a neural network guarantees certain properties with…

机器学习 · 计算机科学 2021-10-04 Kai Jia , Martin Rinard

We investigate the expressive power of neural networks from the point of view of descriptive complexity. We study neural networks that use floating-point numbers and piecewise polynomial activation functions from two perspectives: 1) the…

计算复杂性 · 计算机科学 2025-05-12 Veeti Ahvonen , Damian Heiman , Antti Kuusisto

We study the training of deep neural networks by gradient descent where floating-point arithmetic is used to compute the gradients. In this framework and under realistic assumptions, we demonstrate that it is highly unlikely to find ReLU…

机器学习 · 计算机科学 2023-11-16 Clemens Karner , Vladimir Kazeev , Philipp Christian Petersen

Deep neural network with rectified linear units (ReLU) is getting more and more popular recently. However, the derivatives of the function represented by a ReLU network are not continuous, which limit the usage of ReLU network to situations…

机器学习 · 计算机科学 2020-12-03 Bo Li , Shanshan Tang , Haijun Yu

The synergy between spiking neural networks and neuromorphic hardware holds promise for the development of energy-efficient AI applications. Inspired by this potential, we revisit the foundational aspects to study the capabilities of…

神经与进化计算 · 计算机科学 2024-03-18 Manjot Singh , Adalbert Fono , Gitta Kutyniok

It is commonly recognized that the expressiveness of deep neural networks is contingent upon a range of factors, encompassing their depth, width, and other relevant considerations. Currently, the practical performance of the majority of…

机器学习 · 计算机科学 2023-11-08 Xuan Qi , Yi Wei

We study the memorization power of feedforward ReLU neural networks. We show that such networks can memorize any $N$ points that satisfy a mild separability assumption using $\tilde{O}\left(\sqrt{N}\right)$ parameters. Known VC-dimension…

机器学习 · 计算机科学 2021-10-08 Gal Vardi , Gilad Yehudai , Ohad Shamir

A main open question in contemporary AI research is quantifying the forms of reasoning neural networks can perform when perfectly trained. This paper answers this by interpreting reasoning tasks as circuit emulation, where the gates define…

机器学习 · 计算机科学 2025-09-17 Anastasis Kratsios , Dennis Zvigelsky , Bradd Hart

We discuss the expressive power of neural networks which use the non-smooth ReLU activation function $\varrho(x) = \max\{0,x\}$ by analyzing the approximation theoretic properties of such networks. The existing results mainly fall into two…

泛函分析 · 数学 2019-04-10 Felix Voigtlaender , Philipp Petersen

The celebrated universal approximation theorems for neural networks roughly state that any reasonable function can be arbitrarily well-approximated by a network whose parameters are appropriately chosen real numbers. This paper examines the…

机器学习 · 计算机科学 2023-03-17 C. Sinan Güntürk , Weilin Li

Theoretical studies show that for any differentiable function on a compact domain, there exists a neural network that approximates both the function values and gradients. However, such a result cannot be used in practice since it assumes…

机器学习 · 计算机科学 2026-05-05 Sejun Park , Yeachan Park , Geonho Hwang
‹ 上一页 1 2 3 10 下一页 ›