Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

Convolution: Math, Graphics, and Discrete Signals01:24

Convolution: Math, Graphics, and Discrete Signals

984
In any LTI (Linear Time-Invariant) system, the convolution of two signals is denoted using a convolution operator, assuming all initial conditions are zero. The convolution integral can be divided into two parts: the zero-input or natural response and the zero-state or forced response, with t0 indicating the initial time.
To simplify the convolution integral, it is assumed that both the input signal and impulse response are zero for negative time values. The graphical convolution process...
984
Parallel Processing01:20

Parallel Processing

780
The brain processes sensory information rapidly due to parallel processing, which involves sending data across multiple neural pathways at the same time. This method allows the brain to manage various sensory qualities, such as shapes, colors, movements, and locations, all concurrently. For instance, when observing a forest landscape, the brain simultaneously processes the movement of leaves, the shapes of trees, the depth between them, and the various shades of green. This enables a quick and...
780
Convolution Properties II01:17

Convolution Properties II

600
The important convolution properties include width, area, differentiation, and integration properties.
The width property indicates that if the durations of input signals are T1 and T2, then the width of the output response equals the sum of both durations, irrespective of the shapes of the two functions. For instance, convolving two rectangular pulses with durations of 2 seconds and 1 second results in a function with a width of 3 seconds.
The area property asserts that the area under the...
600
Convolution Properties I01:20

Convolution Properties I

627
Convolution computations can be simplified by utilizing their inherent properties.
The commutative property reveals that the input and the impulse response of an LTI (Linear Time-Invariant) system can be interchanged without affecting the output:
627
Deconvolution01:20

Deconvolution

603
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...
603
Upsampling01:22

Upsampling

646
Managing signal sampling rates is essential in digital signal processing to maintain signal integrity. A decimated signal, characterized by a reduced frequency range due to its lower sampling rate, can be upsampled by inserting zeros between each sample. This upsampling process expands the original spectrum and introduces repeated spectral replicas at intervals dictated by the new Nyquist frequency. To refine this zero-inserted sequence, it is passed through a lowpass filter with a cutoff...
646

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Hydrogen Sulfide-Activated Background-Free Self-Referenced Surface-Enhanced Raman Scattering Nanoprobe for Real-Time and Precise Monitoring of Drug-Induced Liver Injury Progression.

Analytical chemistry·2026
Same author

Integrative Single-Cell and Bulk Transcriptome Analysis Reveals Molecular Mechanisms, Prognostic Signatures, and Therapeutic Response in Glioma.

The Journal of craniofacial surgery·2026
Same author

Metabolic Signaling Meets Epigenetic Regulation: How Protein Lactylation Remodels the Tumor Immune Microenvironment in Gastric Cancer.

Current issues in molecular biology·2026
Same author

A rapid, non-invasive blood-based reporter system for early selection of high-percentage germline chimeras in transgenic chicken production.

Poultry science·2026
Same author

DeepSeek-assisted problem-based learning for glaucoma education in an undergraduate ophthalmology clerkship: a randomized educational pilot study.

Scientific reports·2026
Same author

Effect of Combined Electrowarm Dry Needling on Postpartum Pelvic Girdle Pain: A Randomized Controlled Study.

Journal of visualized experiments : JoVE·2026

Related Experiment Video

Updated: Feb 16, 2026

Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

10.0K

High Performance Implementation of 3D Convolutional Neural Networks on a GPU.

Qiang Lan1,2, Zelong Wang1,2, Mei Wen1,2

  • 1College of Computer, National University of Defense Technology, Changsha 410073, China.

Computational Intelligence and Neuroscience
|December 19, 2017
PubMed
Summary

This study optimizes 3D convolutional neural networks for video classification using the Winograd Minimal Filtering Algorithm (WMFA). The WMFA implementation achieved a twofold speedup in 3D convolution layers compared to cuDNN, enhancing computational efficiency.

Related Experiment Videos

Last Updated: Feb 16, 2026

Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

10.0K

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Machine Learning

Background:

  • Convolutional neural networks (CNNs) excel in 2D tasks but face computational challenges with 3D video data.
  • Existing methods like FFT-based approaches reduce computation but increase memory usage.
  • The Winograd Minimal Filtering Algorithm (WMFA) offers reduced operations and memory for 2D CNNs.

Purpose of the Study:

  • To implement and evaluate the Winograd Minimal Filtering Algorithm (WMFA) for 3D convolutional neural networks (CNNs).
  • To assess the performance of WMFA in accelerating video classification tasks.
  • To compare the WMFA implementation against the cuDNN library for 3D convolutions.

Main Methods:

  • Implementation of the Winograd Minimal Filtering Algorithm (WMFA) specifically for 3D convolutional layers.
  • Application of the WMFA-optimized 3D CNN to a standard video classification network.
  • Comparative performance analysis against the cuDNN library on key 3D convolution operations.

Main Results:

  • The optimized WMFA implementation demonstrated a significant speedup for 3D convolution layers.
  • A twofold speedup was observed in most 3D convolution layers compared to the cuDNN baseline.
  • The WMFA approach successfully accelerated video classification without increasing memory requirements.

Conclusions:

  • The Winograd Minimal Filtering Algorithm (WMFA) is an effective strategy for accelerating 3D convolutional neural networks.
  • This optimization is particularly beneficial for memory-intensive tasks like video classification.
  • The findings suggest WMFA as a viable alternative to existing acceleration methods, offering improved computational efficiency.