arrow
返回

Multi-tailed vision transformer for efficient inference

delete2024-06-01
delete2
delete
OA
AI
Y
Yunke Wang
B
Bo Du *
W
Wenyuan Wang
C
Chang Xu *
DOI:10.1016/j.neunet.2024.106235delete
delete原文链接
delete分享
delete收藏
查看原文
摘要

摘要

En 中文
Recently, Vision Transformer (ViT) has achieved promising performance in image recognition and gradually serves as a powerful backbone in various vision tasks. To satisfy the sequential input of Transformer, the tail of ViT first splits each image into a sequence of visual tokens with a fixed length. Then, the following self -attention layers construct the global relationship between tokens to produce useful representation for the downstream tasks. Empirically, representing the image with more tokens leads to better performance, yet the quadratic computational complexity of self -attention layer to the number of tokens could seriously influence the efficiency of ViT's inference. For computational reduction, a few pruning methods progressively prune uninformative tokens in the Transformer encoder, while leaving the number of tokens before the Transformer untouched. In fact, fewer tokens as the input for the Transformer encoder can directly reduce the following computational cost. In this spirit, we propose a Multi -Tailed Vision Transformer (MT-ViT) in the paper. MT-ViT adopts multiple tails to produce visual sequences of different lengths for the following Transformer encoder. A tail predictor is introduced to decide which tail is the most efficient for the image to produce accurate prediction. Both modules are optimized in an end -to -end fashion, with the Gumbel-Softmax trick. Experiments on ImageNet-1K demonstrate that MT-ViT can achieve a significant reduction on FLOPs with no degradation of the accuracy and outperform compared methods in both accuracy and FLOPs.
Keyword:
Vision transformer
Efficient inference
Dynamic neural network
AI总结

AI总结

对已上传原文的论文进行重点信息的提取,主要内容包括:简要概述、研究摘要、背景介绍、关键亮点、图文解析、展望与总结。

期刊

Neural Networks 封面图
Neural Networks
IF:
6.3
论文数:
8.3K
被引数:
3.0W

机构

U
University of Sydney
学者数:
6.5W
论文数: 6.2W
被引数: 90
W
wuhan university
学者数:
8.1W
论文数: 5.8W
被引数: 70
引用论文

引用论文

err分享
err收藏
Interactive Rendering of Acquired Materials on Dynamic Geometry Using Frequency Analysis
err2013-05-01
err0
errOAAI
errMahdi Mohammad Bagher; Cyril Soler; Kartic Subr; Laurent Belcour; Nicolas Holzschuch
err分享
err收藏
err分享
err收藏
A survey of techniques for optimizing transformer inference变压器推理优化技术综述
err2023-11-01
err31
errOAAI
errChitty-Venkata, Krishna Teja; Mittal, Sparsh; Emani, Murali; Vishwanath, Venkatram; Somani, Arun K.
err分享
err收藏
err分享
err收藏
学者 查看更多内容