相关实验视频

Updated: Jun 24, 2025

Air Filter Devices Including Nonwoven Meshes of Electrospun Recombinant Spider Silk Proteins

Air Filter Devices Including Nonwoven Meshes of Electrospun Recombinant Spider Silk Proteins

Published on: May 8, 2013

基于高效张量分解的过器修剪

Van Tien Pham¹, Yassine Zniyed¹, Thanh Phuong Nguyen¹

¹Université de Toulon, Aix Marseille University, CNRS, LIS UMR 7020, France.

Neural networks : the official journal of the International Neural Network Society

|June 3, 2024

概括

我们介绍了CORING (基于高效张量分解的过器修剪),这是一个新的神经网络修剪方法. CORING使用张量分解来显著降低模型的复杂性,同时在各种视觉任务中保持准确性.

科学领域:

计算机科学计算机科学
人工智能的人工智能
机器学习机器学习

背景情况:

神经网络的修剪对于降低计算成本和记忆足迹至关重要.
传统的修剪方法通常依赖于简化过器表示,可能会丢失信息.
需要有效的修剪技术来保持网络性能.

研究的目的:

引入CORING (基于高效张量分解的过器修剪),这是神经网络的新过器修剪方法.
为了利用张量分解来实现高效和有效的神经网络压缩.
为了证明CORING在现有的最先进的修剪方法上的优势.

主要方法:

CORING利用张量分解,特别是高阶单数值分解 (HOSVD),以在它们的多维形式中近似过器.
引入了一种新的过器相似度指标,基于HOSVD的低级近似值,提高了效率.
该方法在各种神经网络架构和数据集中进行了测试,用于各种计算机视觉任务.

主要成果:

与最先进的方法相比,CORING显著减少了多次积累操作 (MAC) 和参数.
该方法在图像分类,对象检测,实例细分和关键点检测任务中始终提高验证准确性.
除研究和定性结果证实了基于张量方法的效率和保留基本网络特征.

结论:

CORING提供了一种高效和有效的方法,用于使用张量分解进行神经网络过器修剪.
该方法实现了优越的压缩速率,同时提高了模型的准确性.
在深度学习应用程序的模型优化中,CORING 代表了显著的进步.

关键词:

过器的修剪过器的修剪网络压缩网络压缩张量分解的分解方式

更多相关视频

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

Design and Characterization Methodology for Efficient Wide Range Tunable MEMS Filters

Design and Characterization Methodology for Efficient Wide Range Tunable MEMS Filters

Published on: February 4, 2018

相关概念视频

Deconvolution

Deconvolution

Deconvolution, also known as inverse filtering, is the process of extracting the impulse response from known input and output signals. This technique is vital in scenarios where the system's characteristics are unknown, and they must be inferred from the observable signals.
Deconvolution involves several mathematical techniques to derive the impulse response. One common approach is polynomial division. In this method, the input and output sequences are treated as coefficients of...

Filtration

Filtration

Filtration is a physical separation process that involves passing a suspension through a porous medium to separate solids from fluids. During filtration, solids collect on the porous medium while liquids, also collectively known as the filtrate, pass through. The filtration medium is selected based on the filtration purpose, quantity, and nature of the precipitate. The general criteria for a suitable filtering medium are that it is inert, mechanically strong, nonabsorbent toward dissolved...

Fast Fourier Transform

Fast Fourier Transform

The Fast Fourier Transform (FFT) is a computational algorithm designed to compute the Discrete Fourier Transform (DFT) efficiently. By breaking down the calculations into smaller, manageable sections, the FFT significantly reduces the computational complexity involved. Direct computation of an N-point DFT requires N2 complex multiplications, whereas the FFT algorithm needs only (N/2)log⁡2N multiplications, offering a much faster performance.
The computational efficiency of the FFT becomes...

Discrete-time Fourier transform

Discrete-time Fourier transform

The Discrete-Time Fourier Transform (DTFT) is an essential mathematical tool for analyzing discrete-time signals, converting them from the time domain to the frequency domain. This transformation allows for examining the frequency components of discrete signals, providing insights into their spectral characteristics. In the DTFT, the continuous integral used in the continuous-time Fourier transform is replaced by a summation to accommodate the discrete nature of the signal.
One of the notable...

¹³C NMR: Distortionless Enhancement by Polarization Transfer (DEPT)

¹³C NMR: Distortionless Enhancement by Polarization Transfer (DEPT)

When proton-coupled carbon-13 spectra are simplified by a broadband proton decoupling technique, structural information about the coupled protons is lost. Distortionless enhancement by polarization transfer (DEPT) is a technique that provides information on the number of hydrogens attached to each carbon in a molecule. While the DEPT experiment utilizes complex pulse sequences, the pulse delay and flip angle are specifically manipulated. The resulting signals have different phases depending on...

Downsampling

Downsampling

When considering a sampled sequence with zero values between sampling instants, one can replace it by taking every N-th value of the sequence. At these integer multiples of N, the original and sampled sequences coincide. This process, known as decimation, involves extracting every N-th sample from a sequence, thereby creating a more efficient sequence.
The Fourier transform of the decimated sequence reveals a combination of scaled and shifted versions of the original spectrum. This...