logo
科言猫学术研究的AI总结
language
手机端手机端
  • Phone Feature微信小程序
简体中文简体中文
notice消息通知
博客博客
delete
arrow
返回

nnWeb: Towards efficient WebGPU-based DNN inference via automatic collaborative offloading

delete2025-07-01
delete0
PRE
AI
J
Jiawei Liu
B
Bing Dong
王
王维龙 (Weilong Wang)
M
Muhan Yuan
B
Borui Li
Z
Zhao-Dong Xu
S
Shuai Wang
DOI：10.1016/j.comnet.2025.111489delete
delete原文链接
delete原文求助
delete分享
delete收藏
摘要

摘要

En 中文
在浏览器中的神经网络推理为跨平台AI应用带来了前景，但在资源受限设备上面临着严重的延迟和能耗挑战。本文提出nnWeb，一个基于WebGPU的浏览器内神经网络推理框架，具有优化的延迟和能耗效率。nnWeb动态分割神经网络，并促进客户端浏览器与服务器之间的协同卸载。nnWeb运行分为两个阶段：(1)基于层隔离的配置文件分析，用于预测异构硬件上每层的执行延迟和能耗；以及(2)基于异步执行的DNN分割，持续监控网络带宽和设备负载，利用WebGPU的原生流水线并行ism选择最佳分割点，通过运行时求解闭合式优化来最小化总延迟或能耗。在多种浏览器内AI模型和网络条件下的广泛评估表明，与静态分割相比，nnWeb实现了30%至52%的总推理延迟平均减少。此外，与独立的浏览器推理相比，nnWeb实现了11.3%至44.0%的能耗节省。

期刊

Computer Networks 封面图
Computer Networks
IF:
4.6
论文数:
1.6K
被引数:
1.6W

机构

暂无机构信息
引用论文

引用论文

DenseNet-Transformer: A deep learning method for spatial–temporal traffic prediction in optical fronthaul network
err
IF0
err2024-11-01
err0
PREAI
errXin Qin; Wenwu Zhu; Qian Hu; Zexi Zhou; Yi Ding; Xia Gao; Rentao Gu
err分享
err收藏
Finding gene network topologies for given biological function with recurrent neural network
errNATURE COMMUNICATIONS
IF15.7
err2021-05-25
err24
errOAAI
errShen, Jingxiang; Liu, Feng; Tu, Yuhai; Tang, Chao
err分享
err收藏
Edge-Centric Programming for IoT Applications With Automatic Code Partitioning
errIEEE TRANSACTIONS ON COMPUTERS
IF3.8
err2022-10-01
err0
PREAI
errLi, Borui; Dong, Wei
err分享
err收藏
Recent advances in reinforcement learning-based autonomous driving behavior planning: A survey
errTRANSPORTATION RESEARCH PART C-EMERGING TECHNOLOGIES
IF7.9
err2024-07-01
err4
PREAI
errWu, Jingda; Huang, Chao; Huang, Hailong; Lv, Chen; Wang, Yuntong; Wang, Fei-Yue
err分享
err收藏
Neurosurgeon
errACM SIGARCH Computer Architecture News
IF0
err2017-04-04
err0
errOAAI
errYiping Kang; Johann Hauswald; Cao Gao; Austin Rovinski; Trevor Mudge; Jason Mars; Lingjia Tang
err分享
err收藏
A comprehensive analysis of the coverage and performance of 4G and 5G deployments
errComputer Networks
IF0
err2023-12-01
err0
errOAAI
errMuhammad Iqbal Rochman; Vanlin Sathya; Damian Fernandez; Norlen Nunez; Ahmed S. Ibrahim; William Payne; Monisha Ghosh
err分享
err收藏
Model Pruning Enables Efficient Federated Learning on Edge Devices模型修剪可在边缘设备上实现高效的联合学习
errIEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
IF8.9
err2023-12-01
err160
errOAAI
errJiang, Yuang; Wang, Shiqiang; Valls, Victor; Ko, Bong Jun; Lee, Wei-Han; Leung, Kin K.; Tassiulas, Leandros
err分享
err收藏
An enhanced genetic algorithm for computation task offloading in MEC scenario
err
IF0
err2023-01-01
err0
PREAI
errJiacheng Zhao; Wenzao Li; Hantao Liu; Peizhen Yu; Hanyun Li; Zhan Wen
err分享
err收藏
Knacks of a hybrid anomaly detection model using deep auto-encoder driven gated recurrent unit
errComputer Networks
IF0
err2023-05-01
err0
PREAI
errEarum Mushtaq; Aneela Zameer; Rubina Nasir
err分享
err收藏
PowerChop
err
IF0
err2016-06-18
err0
PREAI
errMichael A. Laurenzano; Yunqi Zhang; Jiang Chen; Lingjia Tang; Jason Mars
err分享
err收藏
学者 查看更多内容