Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Related Concept Videos

You might also read

Related Articles

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

Sort by

Same author

Usefulness of the multimodal biopsy approach in radial endobronchial ultrasound biopsy for peripheral lung lesions: a single center retrospective study.

Journal of thoracic disease·2026

Same author

CDK9 degrader induces BRCAness and sensitizes castration-resistant prostate cancer to PARP inhibitor.

Theranostics·2026

Same author

Without Paired Labeled Data: End-to-End Self-Supervised Learning for Drone-View Geo-Localization.

IEEE transactions on neural networks and learning systems·2026

Same author

Fully and spontaneously degradable carboxymethyl chitosan/oxidized hyaluronic acid-based membrane promoting three-dimensional tissue remodeling in vitro and in vivo.

International journal of biological macromolecules·2026

Same author

Tracking the Night: Measuring Age and Sex Patterns in Sleep Duration Using Wearable Technology.

Sleep·2026

Same author

Functionalized polymeric nanomaterials for BMP-2 delivery: recent advances toward improved bone regeneration.

Nanomedicine (London, England)·2026

Search research articles

Related Experiment Video

Updated: Nov 1, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

QTTNet: Quantized tensor train neural networks for 3D object and video recognition.

Donghyun Lee¹, Dingheng Wang², Yukuan Yang¹

¹Department of Precision Instrumentation, Center for Brain Inspired Computing Research and Beijing Innovation Center for Future Chip, Tsinghua University, Beijing 100084, China.

Neural Networks : the Official Journal of the International Neural Network Society

|June 19, 2021

Summary

We developed QTTNet, a framework combining tensor train decomposition and data quantization to significantly shrink 3D convolutional neural networks (3DCNNs). This reduces memory and time costs for resource-limited devices with minimal accuracy loss.

Keywords:

3DCNN 8 bit inference Neural network compression Quantization Tensor train decomposition

More Related Videos

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

Detection and Quantification of Tunneling Nanotubes Using 3D Volume View Images

Detection and Quantification of Tunneling Nanotubes Using 3D Volume View Images

Published on: August 31, 2022

Related Experiment Videos

Last Updated: Nov 1, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

Detection and Quantification of Tunneling Nanotubes Using 3D Volume View Images

Detection and Quantification of Tunneling Nanotubes Using 3D Volume View Images

Published on: August 31, 2022

Area of Science:

Computer Vision
Machine Learning
Deep Learning

Background:

Convolutional Neural Networks (CNNs) achieve state-of-the-art results but incur high memory costs.
Large CNN models are challenging to deploy on resource-limited portable devices.

Purpose of the Study:

To introduce QTTNet, a novel training framework for 3DCNNs.
To significantly reduce the model size and computational cost of 3DCNNs for efficient deployment.

Main Methods:

QTTNet integrates Tensor Train (TT) decomposition with data quantization.
TT decomposition reduces trainable parameters, while quantization decreases data bit-width.
Low bit quantization is applied to TT-cores, activations, and batch normalizations during inference.

Main Results:

QTTNet achieves substantial compression ratios, up to nearly 180x, with minimal accuracy degradation.
Models exhibit significantly lower memory and time costs compared to conventional methods.
Experimental validation on ModelNet40, UCF11, and UCF50 datasets demonstrates effectiveness for 3D object and video recognition.

Conclusions:

QTTNet offers an effective solution for compressing 3DCNNs, enabling their use on portable devices.
The framework successfully balances model compression with performance.
QTTNet models show a remarkable reduction in storage requirements, up to 1000x lower than some existing practices.