Concatenated Attention: A Novel Method for Regulating Information Structure Based on Sensors

This paper addresses the challenges of limited training data and suboptimal environmental conditions in image processing tasks, such as underwater imaging with poor lighting and distortion. Neural networks, including Convolutional Neural Networks (CNNs) and Transformers, have advanced image analysis...

Full description

Saved in:
Bibliographic Details
Main Authors: Zeyu Zhang, Tianqi Chen, Yuki Todo
Format: Article
Language:English
Published: MDPI AG 2025-01-01
Series:Applied Sciences
Subjects:
Online Access:https://www.mdpi.com/2076-3417/15/2/523
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1832589283136897024
author Zeyu Zhang
Tianqi Chen
Yuki Todo
author_facet Zeyu Zhang
Tianqi Chen
Yuki Todo
author_sort Zeyu Zhang
collection DOAJ
description This paper addresses the challenges of limited training data and suboptimal environmental conditions in image processing tasks, such as underwater imaging with poor lighting and distortion. Neural networks, including Convolutional Neural Networks (CNNs) and Transformers, have advanced image analysis but remain constrained by computational demands and insufficient data. To overcome these limitations, we propose a novel split-and-concatenate method for self-attention mechanisms. By splitting Query and Key matrices into submatrices, performing cross-multiplications, and applying weighted summation, the method optimizes intermediate variables without increasing computational costs. Experiments on a real-world crack dataset demonstrate its effectiveness in improving network performance.
format Article
id doaj-art-ad701b6993da414eb9c1b562de569fd3
institution Kabale University
issn 2076-3417
language English
publishDate 2025-01-01
publisher MDPI AG
record_format Article
series Applied Sciences
spelling doaj-art-ad701b6993da414eb9c1b562de569fd32025-01-24T13:19:41ZengMDPI AGApplied Sciences2076-34172025-01-0115252310.3390/app15020523Concatenated Attention: A Novel Method for Regulating Information Structure Based on SensorsZeyu Zhang0Tianqi Chen1Yuki Todo2Division of Electrical Engineering and Computer Science, Kanazawa University, Kanazawa 9201192, JapanDivision of Electrical Engineering and Computer Science, Kanazawa University, Kanazawa 9201192, JapanFaculty of Electrical, Information and Communication Engineering, Kanazawa University, Kanazawa 9201192, JapanThis paper addresses the challenges of limited training data and suboptimal environmental conditions in image processing tasks, such as underwater imaging with poor lighting and distortion. Neural networks, including Convolutional Neural Networks (CNNs) and Transformers, have advanced image analysis but remain constrained by computational demands and insufficient data. To overcome these limitations, we propose a novel split-and-concatenate method for self-attention mechanisms. By splitting Query and Key matrices into submatrices, performing cross-multiplications, and applying weighted summation, the method optimizes intermediate variables without increasing computational costs. Experiments on a real-world crack dataset demonstrate its effectiveness in improving network performance.https://www.mdpi.com/2076-3417/15/2/523self-attentiontransformerneural networksensor
spellingShingle Zeyu Zhang
Tianqi Chen
Yuki Todo
Concatenated Attention: A Novel Method for Regulating Information Structure Based on Sensors
Applied Sciences
self-attention
transformer
neural network
sensor
title Concatenated Attention: A Novel Method for Regulating Information Structure Based on Sensors
title_full Concatenated Attention: A Novel Method for Regulating Information Structure Based on Sensors
title_fullStr Concatenated Attention: A Novel Method for Regulating Information Structure Based on Sensors
title_full_unstemmed Concatenated Attention: A Novel Method for Regulating Information Structure Based on Sensors
title_short Concatenated Attention: A Novel Method for Regulating Information Structure Based on Sensors
title_sort concatenated attention a novel method for regulating information structure based on sensors
topic self-attention
transformer
neural network
sensor
url https://www.mdpi.com/2076-3417/15/2/523
work_keys_str_mv AT zeyuzhang concatenatedattentionanovelmethodforregulatinginformationstructurebasedonsensors
AT tianqichen concatenatedattentionanovelmethodforregulatinginformationstructurebasedonsensors
AT yukitodo concatenatedattentionanovelmethodforregulatinginformationstructurebasedonsensors