Expert Systems with Applications最新文献

筛选
英文 中文
A surrogate-assisted evolutionary algorithm with solution sets classification based on inter-dimensional correlation and its applications 基于维间关联的求解集分类代理辅助进化算法及其应用
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-19 DOI: 10.1016/j.eswa.2025.128177
Rui Zhang , Zehua Dong , Chaoli Sun , Yanjun Zhang , Xiaolu Bai
{"title":"A surrogate-assisted evolutionary algorithm with solution sets classification based on inter-dimensional correlation and its applications","authors":"Rui Zhang ,&nbsp;Zehua Dong ,&nbsp;Chaoli Sun ,&nbsp;Yanjun Zhang ,&nbsp;Xiaolu Bai","doi":"10.1016/j.eswa.2025.128177","DOIUrl":"10.1016/j.eswa.2025.128177","url":null,"abstract":"<div><div>To effectively mine the complex long-term dependencies correlations between high-dimensional decision variables, improve the quality of the candidate and real solution sets for evaluation, and expedite the efficiency of fitting the objective function, the paper proposes an algorithm called a surrogate-assisted evolutionary algorithm (SAEA) with solution sets classification based on inter-dimensional correlation for expensive multi-objective optimization (DCSCSAEA) in this study. The paper develops a surrogate model with inter-dimensional correlation called DCBiLSTM, which can carry out nonlinear fitting at a low computational cost, to mine the long-term dependencies correlations between high-dimensional decision variables. A step classification axis is designed based on the reference solutions screened by the division of the radial space, predicting the dominant relationship between the solutions in the space and the reference solutions on the classification axis, in order to classify solutions in the space of the solution set. The paper then divides the uncertainty in the prediction space and develop an adaptive “checkers” model infill criterion to determine the interval in which the predictive error falls. The paper uses the results to choose the corresponding strategy for screening the set of candidate solutions. Experiments on expensive multi-objective optimization problems (EMOPs) (20–100 variables) show DCSCSAEA outperforms five state-of-the-art SAEAs, yielding well-converged, diverse solutions. In real-world weld defect detection (21 variables), DCSCSAEA optimizes the network faster, reducing computational complexity and detection time by 52.14% and 20.39% respectively while maintaining comparable accuracy to state-of-the-art SAEAs.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"287 ","pages":"Article 128177"},"PeriodicalIF":7.5,"publicationDate":"2025-05-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144084208","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Knowledge enhanced and guided graph contrastive learning for molecular property prediction 知识增强和引导图对比学习在分子性质预测中的应用
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-19 DOI: 10.1016/j.eswa.2025.128190
Kunjie Dong, Yanhui Zhang, Xiaohui Lin
{"title":"Knowledge enhanced and guided graph contrastive learning for molecular property prediction","authors":"Kunjie Dong,&nbsp;Yanhui Zhang,&nbsp;Xiaohui Lin","doi":"10.1016/j.eswa.2025.128190","DOIUrl":"10.1016/j.eswa.2025.128190","url":null,"abstract":"<div><div>Molecular property prediction (MPP) lies at the core of fundamental tasks for AI-aided drug discovery. Recent studies have shown great promise in applying self-supervised learning (SSL) to cope with the data scarcity in MPP, and contrastive learning is one of the mainstream methods in SSL. However, current molecular graph contrastive learning methods suffer from two main challenges: molecular graph augmentation that preserves the molecular chemical semantics, and contrastive goal that captures the precise prior knowledge. To address these issues, we propose the <u>K</u>nowledge <u>E</u>nhanced and <u>G</u>uided <u>G</u>raph <u>C</u>ontrastive <u>L</u>earning (KEGGCL). KEGGCL adopts the chemical element domain knowledge to generate two enhanced molecular graphs without changing the molecular chemical structure, ensuring the preservation of the molecular semantics and structure. Next, KEGGCL uses the quantitative estimate of drug-likeness as the guideline to push away sample pairs constituted of different molecules discriminately, capturing the precise prior knowledge. Then, KEGGCL utilizes the well-trained encoders on the featured molecular graph and two element knowledge enhanced molecular graphs to decide the final prediction jointly. Experiments on the 10 benchmark datasets from MoleculeNet show the superiority of KEGGCL. It provides a new graph contrastive manner to learn the precise prior knowledge for better predicting molecular property.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128190"},"PeriodicalIF":7.5,"publicationDate":"2025-05-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144138873","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Graph-based technology recommendation system using GAT-NGCF 基于图技术的GAT-NGCF推荐系统
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-19 DOI: 10.1016/j.eswa.2025.128240
Min-Seung Kim , Yong-Ju Jang , Tae-Eung Sung
{"title":"Graph-based technology recommendation system using GAT-NGCF","authors":"Min-Seung Kim ,&nbsp;Yong-Ju Jang ,&nbsp;Tae-Eung Sung","doi":"10.1016/j.eswa.2025.128240","DOIUrl":"10.1016/j.eswa.2025.128240","url":null,"abstract":"<div><div>This study proposes a GAT-NGCF-based technology recommendation system to improve firms’ technological innovation capabilities and facilitate technology transfer. The system leverages Graph Attention Networks (GAT) to generate optimal representations of firms and patents, which are then applied in a firm–patent interaction graph using Neural Graph Collaborative Filtering (NGCF) to recommend the most suitable patents for transfer. Experiments conducted on 6,797 technology transfer and valuation cases demonstrated high performance, achieving a Recall@5 of 0.9984 and NDCG@5 of 0.9972. Notably, the proposed system outperformed State-Of-The-Art (SOTA) models in collaborative filtering, reinforcing its effectiveness. The system offers customized technology recommendations that align with firms’ technological needs and is expected to play a key role in supporting technology transfer and commercialization strategies through open innovation.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128240"},"PeriodicalIF":7.5,"publicationDate":"2025-05-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144131009","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Customizable multimodal trajectory prediction via nodes of interest selection for autonomous vehicles 基于兴趣节点选择的自定义多模态轨迹预测
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-18 DOI: 10.1016/j.eswa.2025.128222
Titong Jiang, Qing Dong, Yuan Ma, Xuewu Ji, Yahui Liu
{"title":"Customizable multimodal trajectory prediction via nodes of interest selection for autonomous vehicles","authors":"Titong Jiang,&nbsp;Qing Dong,&nbsp;Yuan Ma,&nbsp;Xuewu Ji,&nbsp;Yahui Liu","doi":"10.1016/j.eswa.2025.128222","DOIUrl":"10.1016/j.eswa.2025.128222","url":null,"abstract":"<div><div>To safely navigate through complex traffic scenarios, autonomous vehicles (AVs) must accurately predict the future trajectories of surrounding agents. Therefore, there has been a surge of interest in the problem of trajectory prediction for AVs. Building upon existing studies, we aim to push the boundaries of state-of-the-art research by tackling the following challenges: (1) the interaction between agents is heavily dependent on road geometry and topology; (2) certain modalities of the surrounding agent are non-informative for the AV and can be disregarded; and (3) the diversity of multimodal prediction is limited by the maximum number of modalities. In this study, we propose Customizable Multimodal Transformer (CMT), a deep learning model which facilitates customizable multimodal trajectory prediction. First, inspired by the dependency between agent interaction and road geometry and topology, we propose that map information can be utilized to better understand agent interaction. Furthermore, we propose the concept of nodes of interest (NOI), which represents the area of interest of the AV. By manipulating the nodes in the NOI, CMT can generate customized prediction results where irrelevant modalities can be disregarded without compromising the safety of the AV, leading to reduced computational costs. Finally, we propose to enhance the diversity of multimodal prediction results through Gaussian mixture reduction via clustering (GMRC). Extensive experiments on nuScenes and Argoverse datasets demonstrate that CMT not only outperforms previous state-of-the-art models, but also exhibits great potential for reducing computational costs and improving inference speed for trajectory prediction of AVs. Code is available at https://github.com/Promisery/CMT.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128222"},"PeriodicalIF":7.5,"publicationDate":"2025-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144138870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Pearson and intra-inter-class weighted block diagonal representation learning for subspace clustering 子空间聚类的Pearson和类间加权块对角表示学习
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-18 DOI: 10.1016/j.eswa.2025.128185
Xuemei Zhao , Yusong Xiong , Chen Li , Jun Wu , Qi Zhang , Haijian Wang
{"title":"Pearson and intra-inter-class weighted block diagonal representation learning for subspace clustering","authors":"Xuemei Zhao ,&nbsp;Yusong Xiong ,&nbsp;Chen Li ,&nbsp;Jun Wu ,&nbsp;Qi Zhang ,&nbsp;Haijian Wang","doi":"10.1016/j.eswa.2025.128185","DOIUrl":"10.1016/j.eswa.2025.128185","url":null,"abstract":"<div><div>Block diagonal is an important characteristic of the self-expression coefficient matrix in subspace clustering. However, the global constraint on the self-expression coefficient matrix suffers from the impact of inter-class similarity and the intra-class dissimilarity, as well as the disturbance of noise. To facilitate subspace clustering by enhancing the representation ability of the self-expression coefficient matrix, we propose an enhanced block diagonal representation(BDR) learning that considers internal and external data correlations from the perspectives of pairwise and classwise correlation. First, Pearson correlation is employed to describe local pairwise similarities and acts as a weight to strengthen the corresponding connections between pairwise data points in the self-expression coefficient matrix. Then, a unified intra-class and inter-class dissimilarity constraint is proposed to increase the coefficient values of the same class and decrease the coefficient values of different classes, in other words, enhance intra-class compactness and inter-class separability at the classwise level. In this way, a multi-level constraint on the self-expression coefficient matrix is proposed, from pairwise to classwise along with the global-wise BDR constraint. Experimental results show that the block diagonal structure of the self-expression coefficient matrix is significantly improved with these two additional constraints. Further, with the enhanced self-expression coefficient matrix, the accuracies of the clustering results are also improved.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128185"},"PeriodicalIF":7.5,"publicationDate":"2025-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144154942","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Comparative analysis of multi-criteria decision making methods for prioritizing influential factors of ChatGPT adoption in higher education 高等教育采用ChatGPT影响因素排序的多准则决策方法比较分析
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-18 DOI: 10.1016/j.eswa.2025.128188
Attasit Wiangkham , Rattawut Vongvit
{"title":"Comparative analysis of multi-criteria decision making methods for prioritizing influential factors of ChatGPT adoption in higher education","authors":"Attasit Wiangkham ,&nbsp;Rattawut Vongvit","doi":"10.1016/j.eswa.2025.128188","DOIUrl":"10.1016/j.eswa.2025.128188","url":null,"abstract":"<div><div>Various Multi-Criteria Decision-Making (MCDM) methods are analyzed to identify key factors influencing the adoption of ChatGPT in higher education. The research focuses on understanding the factors affecting ChatGPT adoption in educational settings, to provide a comprehensive framework to evaluate and rank these factors, including usage, agent, technical, and trust-related considerations. To achieve this, the study employs techniques such as the Weighted Sum Model (WSM), Shapley Additive Explanations (SHAP), and Local Interpretable Model-agnostic Explanations (LIME). The methodology involves surveying 55 students from a mechanical engineering program, initially using conventional research methods and later integrating ChatGPT into their workflow. The results, derived from comparing both approaches, provide insights into the practical implications of integrating ChatGPT in education. The findings highlight the importance of a user-friendly interface, multimodal communication, security, emphasizing the need for accessible, intuitive platforms to promote adoption and engagement. Educational institutions should focus on simplifying navigation and enhancing accessibility, ensuring that both students and educators, especially those with limited technical experience, can easily interact with ChatGPT. This study offers valuable insights into the complexities of ChatGPT adoption, helping institutions develop customized strategies, effective implementation strategies. The results suggest that integrating ChatGPT into educational settings brings significant benefits, underscoring the need for a strategic, context-specific approach to its adoption.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"287 ","pages":"Article 128188"},"PeriodicalIF":7.5,"publicationDate":"2025-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144117138","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
A novel biologically plausible spiking convolutional capsule network with optimized batch normalization for EEG-based emotion recognition 一种具有优化批归一化的基于脑电图的情感识别的新型生物似然尖峰卷积胶囊网络
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-18 DOI: 10.1016/j.eswa.2025.128183
Kun Chen , Zihao Yang , Mincheng Cai , Quan Liu , Qingsong Ai , Li Ma
{"title":"A novel biologically plausible spiking convolutional capsule network with optimized batch normalization for EEG-based emotion recognition","authors":"Kun Chen ,&nbsp;Zihao Yang ,&nbsp;Mincheng Cai ,&nbsp;Quan Liu ,&nbsp;Qingsong Ai ,&nbsp;Li Ma","doi":"10.1016/j.eswa.2025.128183","DOIUrl":"10.1016/j.eswa.2025.128183","url":null,"abstract":"<div><div>Emotion recognition based on Electroencephalogram (EEG) is currently a hot topic in human-computer interaction, as EEG can more accurately reflect the characteristics of emotion. In recent years, EEG emotion recognition based on deep learning has achieved significant progress, and particularly methods combined with a capsule network (CapsNet) have outstanding performance. However, as the complexity of models continues to increase, resource consumption has also escalated. In this context, spiking neural networks (SNNs), known for their energy efficiency and greater biological plausibility, have attracted attentions of numerous researchers. Nevertheless, convolutional neural networks (CNNs) -based methods are not effective in SNN, and there remains a considerable gap compared with artificial neural networks (ANNs). This paper proposes a novel method combining the high-performing capsule networks with SNNs, named the spiking convolutional capsule network (SCCapsNet) for EEG-based emotion recognition tasks. To our knowledge, this is the first attempt to introduce capsule networks into SNNs for EEG emotion recognition. Furthermore, the spike-timing-dependent plasticity (STDP) routing algorithm is improved to sensitively capture temporal sequence information of EEG signals to enhance biological plausibility of SCCapsNet. In addition, a novel batch normalization (BN) layer incorporating the membrane potential decay time constant (tau-BN) is suggested to address the issue of neuron death caused by reduction in spike firing rate due to the <span><math><msub><mi>L</mi><mn>2</mn></msub></math></span> norm. We provide a theoretical explanation of the role of the BN layer in regulating spike firing rates. Finally, the performance of SCCapsNet is validated on two public datasets. As for DEAP dataset, SCCapsNet achieved recognition accuracies of 97.01 %, 96.84 %, and 96.73 % for valence, arousal, and dominance dimensions, respectively. The accuracies are 89.82 %, 93.69 %, and 93.90 % on the same dimensions for DREAMER dataset. A recognition accuracy of 90.32 % was achieved on the five-category dataset SEED-V. Experimental results outperform all other comparable SNN methods. In addition, we validated the enhancing effect of the proposed tau-BN on spike firing rates. The results showed that the enhancement effect was obvious, successfully addressing the issue of neuron death caused by excessively low spike firing rates due to the <span><math><msub><mi>L</mi><mn>2</mn></msub></math></span> norm. Our code is available at <span><span>https://github.com/Zihao0/SCCapsNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128183"},"PeriodicalIF":7.5,"publicationDate":"2025-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124949","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
FMFI: Transformer based four branches multi-granularity feature integration for person Re-ID FMFI:基于转换器的四分支多粒度特性集成,用于人员Re-ID
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-18 DOI: 10.1016/j.eswa.2025.128150
Xiaohan Zheng, Jian Lu, Jiahui Xing, Kaibing Zhang
{"title":"FMFI: Transformer based four branches multi-granularity feature integration for person Re-ID","authors":"Xiaohan Zheng,&nbsp;Jian Lu,&nbsp;Jiahui Xing,&nbsp;Kaibing Zhang","doi":"10.1016/j.eswa.2025.128150","DOIUrl":"10.1016/j.eswa.2025.128150","url":null,"abstract":"<div><div>Extracting multi-granularity features is a critical challenge in person re-identification (Re-ID). While convolutional neural networks (CNNs) are effectively capture local salient features through convolutional kernels, they struggle to construct globally discriminative representations. In contrast, Transformer networks can model long-range dependencies and establish global contextual relationships among features, making them a powerful tool for multi-granularity feature learning in Re-ID. To comprehensively extract multi-granularity features from various aspects such as clothing attributes, walking postures, and social interactions, we propose FMFI, a four-branch multi-granularity feature extraction and integration method based on Transformer. FMFI employs a four-branch architecture to capture diverse feature representations, enhancing the model’s expressiveness and robustness. Specifically, we introduce the Quadra-Net (QN) structure, which extends from the final layer of a replicated Transformer. By appropriately scaling the branch-wise feature weights and aggregating global tokens from all four branches, FMFI constructs enhanced global feature representations. Furthermore, we design the Refined Global Feature (RGF) module, which refines the initial global features and establishes connections with the newly integrated features, leading to more distinctive and discriminative global representations. Extensive experiments on the Market1501, CUHK03, and MSMT17 Re-ID datasets demonstrate that the proposed FMFI method outperforms most existing Re-ID approaches. Our model significantly enriches feature representations and improves the extraction of multi-granularity features, thereby enhancing person re-identification performance.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128150"},"PeriodicalIF":7.5,"publicationDate":"2025-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144131007","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Deep Shallow network with LSTM for detecting attacks in IoT networks and preserving privacy based on Adaptive hybrid encryption algorithm 基于LSTM的深浅网络检测物联网网络中的攻击,并基于自适应混合加密算法保护隐私
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-18 DOI: 10.1016/j.eswa.2025.128050
Deepak Dilip Mahajan , A. Jeyasekar
{"title":"Deep Shallow network with LSTM for detecting attacks in IoT networks and preserving privacy based on Adaptive hybrid encryption algorithm","authors":"Deepak Dilip Mahajan ,&nbsp;A. Jeyasekar","doi":"10.1016/j.eswa.2025.128050","DOIUrl":"10.1016/j.eswa.2025.128050","url":null,"abstract":"<div><div>In this article, a new attack detection and privacy preservation framework is implemented to identify the attacks present in IoT networks and preserve the information from various attacks during transmission. Initially, the optimal features are selected from the garnered data by using the Adaptive Random Index-based Sea Lion Optimization Algorithm (ARI-SLOA). Consequently, the resultant features are given to a Deep Shallow Network with Long Short-Term Memory (DSN-LSTM) for attack identification. The attacks present in the IoT are mitigated during data transmission, and thus the data is highly secured. The designed encryption scheme is implemented for preserving the privacy of the information, where Adaptive Hybrid Attribute-based Encryption with an Advanced Encryption System (AHABE-AES) is utilized for privacy preservation. Here, the parameters of the AHABE-AES are optimized by the ARI-SLOA. The developed framework’s performance is examined with other existing approaches. From the results, the suggested framework obtained 95% accuracy, 8% FDR and 4% FNR rates. The outcomes obtained from the developed system ensure that this designed strategy is more robust and effective than other related models.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128050"},"PeriodicalIF":7.5,"publicationDate":"2025-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144125277","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
KFF: K-feature fusion token merging for vision transformer KFF:用于视觉转换器的k特征融合令牌合并
IF 7.5 1区 计算机科学
Expert Systems with Applications Pub Date : 2025-05-18 DOI: 10.1016/j.eswa.2025.128206
Yu Yang , Yue Zhou , Xiaofang Hu , Shukai Duan
{"title":"KFF: K-feature fusion token merging for vision transformer","authors":"Yu Yang ,&nbsp;Yue Zhou ,&nbsp;Xiaofang Hu ,&nbsp;Shukai Duan","doi":"10.1016/j.eswa.2025.128206","DOIUrl":"10.1016/j.eswa.2025.128206","url":null,"abstract":"<div><div>Recently, Vision Transformer (ViT) has achieved better performance than Convolutional Neural Networks (CNNs) in various vision applications. However, they are usually more computationally expensive than CNNs and face challenges in training and inference efficiency. Token merging is an effective and training-free way to reduce model complexity. However, since few tokens are exactly the same, prevalent similarity-based merging methods are challenging to avoid feature information loss and accuracy degradation. To address this issue, we propose a novel K-feature fusion token merging algorithm that significantly reduces the similarity metric error and token merging error with almost no accuracy loss. Specifically, we first reveal that similarity measurement errors and merging strategies have a significant impact on the performance of token merging algorithms, and the currently popular K-based similarity method will cause obvious feature shifts during the merging process. Based on this observation, we present a new feature-enhanced K-feature fusion token similarity calculation method. By combining the keys (K), which summarize the information contained in each token, and the more detailed intermediate features, the error of similarity measurement is greatly reduced. Then, we design a similarity-weighted average token merging algorithm to combine tokens that is faster and more accurate than ordinary average token merging. Extensive experiments show that our approach yields better model performance when reducing comparable computational effort and improving throughput without extra training. For example, for ViT-B on ImageNet, our method reduces 49.58 % of tokens and improves throughput by 30 % with only a 0.44 % drop in accuracy.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"288 ","pages":"Article 128206"},"PeriodicalIF":7.5,"publicationDate":"2025-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144138872","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
相关产品
×
本文献相关产品
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信