Expert Systems with Applications最新文献_第9页

A multi-task minutiae transformer network for fingerprint recognition of young children

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126825

Manhua Liu , Aitong Liu , Yelin Shi , Shuxin Liu

Fingerprint recognition of children have attracted increasing attention for real applications such as identity certificate. However, the recognition performance is greatly reduced if the existing systems are directly used on the fingerprints of young children due to their low resolution and poor image quality. Towards more accurate fingerprint recognition of young children, this paper proposes multi-task deep learning framework based on Pyramid Densely-connected U-shaped Swin-transformer network (PDUSwin-Net) to jointly learn the reconstruction of enhanced high-resolution images and detection of minutiae points, which is compatible with existing adult fingerprint sensors (500 dpi) and minutiae matchers. First, a pyramid densely-connected U-shaped convolutional network is proposed to learn the features of fingerprints for multiple tasks. Then, a swin-transformer attention block is added to model the correlations of long-spatial features. In the decoding part, two branches are built for the tasks of fingerprint enhancement and minutiae extraction. Finally, our method is tested with the existing matchers on two independent fingerprint datasets of young children aged from 0–2 years. Results and comparison show that our method performs better than other methods for fingerprint recognition of young children.

{"title":"A multi-task minutiae transformer network for fingerprint recognition of young children","authors":"Manhua Liu , Aitong Liu , Yelin Shi , Shuxin Liu","doi":"10.1016/j.eswa.2025.126825","DOIUrl":"10.1016/j.eswa.2025.126825","url":null,"abstract":"<div><div>Fingerprint recognition of children have attracted increasing attention for real applications such as identity certificate. However, the recognition performance is greatly reduced if the existing systems are directly used on the fingerprints of young children due to their low resolution and poor image quality. Towards more accurate fingerprint recognition of young children, this paper proposes multi-task deep learning framework based on Pyramid Densely-connected U-shaped Swin-transformer network (PDUSwin-Net) to jointly learn the reconstruction of enhanced high-resolution images and detection of minutiae points, which is compatible with existing adult fingerprint sensors (500 dpi) and minutiae matchers. First, a pyramid densely-connected U-shaped convolutional network is proposed to learn the features of fingerprints for multiple tasks. Then, a swin-transformer attention block is added to model the correlations of long-spatial features. In the decoding part, two branches are built for the tasks of fingerprint enhancement and minutiae extraction. Finally, our method is tested with the existing matchers on two independent fingerprint datasets of young children aged from 0–2 years. Results and comparison show that our method performs better than other methods for fingerprint recognition of young children.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126825"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143446056","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

Recommendation of TV programs via information filtering in RCA tripartite networks

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126836

Kean Li , An Zeng , Jianlin Zhou , Yijun Chen , Xiaohua Cui

Television remains an indispensable medium for information and entertainment, even in the era of widespread streaming media. With the expansion of TV channels through set-top boxes, users now face an overwhelming variety of choices, leading to information overload problems. Recommendation systems have effectively solved the information overload problem and can thus be naturally applied to television. Prior research has focused on improvements in algorithms and the addition of other data. In this paper, without introducing external data, we generate recommendations based on 3 months of TV viewing data from a Chinese city. Considering the large amount of noisy data caused by short stays in TV programs, we simplify the original almost fully connected tripartite network by eliminating the insignificant links with the Revealed Comparative Advantage (RCA) metric to comprehensively reflect user preferences. The inclusion of channel nodes allows the network structure to better align with user behavior characteristics, which differs from traditional bipartite networks that only include user-program interactions. We examine data with different sparsity and find that our approach continues to outperform conventional bipartite network recommendations in terms of accuracy. The advantages of our approach have been validated through comparisons with other advanced methods and across different datasets. Overall, only based on viewing records of users, our work provides accurate TV program recommendations that can capture the underlying user behavior characteristics.

{"title":"Recommendation of TV programs via information filtering in RCA tripartite networks","authors":"Kean Li , An Zeng , Jianlin Zhou , Yijun Chen , Xiaohua Cui","doi":"10.1016/j.eswa.2025.126836","DOIUrl":"10.1016/j.eswa.2025.126836","url":null,"abstract":"<div><div>Television remains an indispensable medium for information and entertainment, even in the era of widespread streaming media. With the expansion of TV channels through set-top boxes, users now face an overwhelming variety of choices, leading to information overload problems. Recommendation systems have effectively solved the information overload problem and can thus be naturally applied to television. Prior research has focused on improvements in algorithms and the addition of other data. In this paper, without introducing external data, we generate recommendations based on 3 months of TV viewing data from a Chinese city. Considering the large amount of noisy data caused by short stays in TV programs, we simplify the original almost fully connected tripartite network by eliminating the insignificant links with the Revealed Comparative Advantage (RCA) metric to comprehensively reflect user preferences. The inclusion of channel nodes allows the network structure to better align with user behavior characteristics, which differs from traditional bipartite networks that only include user-program interactions. We examine data with different sparsity and find that our approach continues to outperform conventional bipartite network recommendations in terms of accuracy. The advantages of our approach have been validated through comparisons with other advanced methods and across different datasets. Overall, only based on viewing records of users, our work provides accurate TV program recommendations that can capture the underlying user behavior characteristics.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126836"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143453667","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

Stockformer: A price–volume factor stock selection model based on wavelet transform and multi-task self-attention networks

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126803

Bohan Ma , Yushan Xue , Yuan Lu, Jing Chen

As the Chinese stock market continues to evolve and its market structure grows increasingly complex, traditional quantitative trading methods face escalating challenges. Due to policy uncertainty and frequent market fluctuations triggered by sudden economic events, existing models often struggle to predict market dynamics accurately. To address these challenges, this paper introduces “Stockformer,” a price–volume factor stock selection model that integrates wavelet transformation and a multitask self-attention network to enhance responsiveness and predictive accuracy regarding market instabilities. Through discrete wavelet transform, Stockformer decomposes stock returns into high and low frequencies, meticulously capturing long-term market trends and short-term fluctuations, including abrupt events. Moreover, the model incorporates a Dual-Frequency Spatiotemporal Encoder and graph embedding techniques to capture complex temporal and spatial relationships among stocks effectively. Employing a multitask learning strategy, it simultaneously predicts stock returns and directional trends. Experimental results show that Stockformer outperforms existing advanced methods on multiple real stock market datasets. In strategy backtesting, Stockformer consistently demonstrates exceptional stability and reliability across market conditions—whether rising, falling, or fluctuating—particularly maintaining high performance during downturns or volatile periods, indicating high adaptability to market fluctuations. To foster innovation and collaboration in the financial analysis sector, the Stockformer model’s code has been open-sourced and is available on the GitHub repository: https://github.com/Eric991005/Multitask-Stockformer.

{"title":"Stockformer: A price–volume factor stock selection model based on wavelet transform and multi-task self-attention networks","authors":"Bohan Ma , Yushan Xue , Yuan Lu, Jing Chen","doi":"10.1016/j.eswa.2025.126803","DOIUrl":"10.1016/j.eswa.2025.126803","url":null,"abstract":"<div><div>As the Chinese stock market continues to evolve and its market structure grows increasingly complex, traditional quantitative trading methods face escalating challenges. Due to policy uncertainty and frequent market fluctuations triggered by sudden economic events, existing models often struggle to predict market dynamics accurately. To address these challenges, this paper introduces “Stockformer,” a price–volume factor stock selection model that integrates wavelet transformation and a multitask self-attention network to enhance responsiveness and predictive accuracy regarding market instabilities. Through discrete wavelet transform, Stockformer decomposes stock returns into high and low frequencies, meticulously capturing long-term market trends and short-term fluctuations, including abrupt events. Moreover, the model incorporates a Dual-Frequency Spatiotemporal Encoder and graph embedding techniques to capture complex temporal and spatial relationships among stocks effectively. Employing a multitask learning strategy, it simultaneously predicts stock returns and directional trends. Experimental results show that Stockformer outperforms existing advanced methods on multiple real stock market datasets. In strategy backtesting, Stockformer consistently demonstrates exceptional stability and reliability across market conditions—whether rising, falling, or fluctuating—particularly maintaining high performance during downturns or volatile periods, indicating high adaptability to market fluctuations. To foster innovation and collaboration in the financial analysis sector, the Stockformer model’s code has been open-sourced and is available on the GitHub repository: <span><span>https://github.com/Eric991005/Multitask-Stockformer</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126803"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143428992","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

Dynamic programming-based exact and heuristic algorithms for single machine scheduling with sequence-dependent setups

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126866

Tengmu Hu , Shih-Hsien Tseng , Theodore T. Allen

This study presents a novel algorithmic framework and an inventory flow mixed integer programming formulation designed to minimize total tardiness and the number of setups. The approach decomposes the problem into three stages: intra-family scheduling, family sequence optimization, and family-switch timing. We propose a specialized heuristic with

O (n^{5} log n)

complexity efficiently handles intra-family scheduling and is extended to accommodate subfamily groupings. Dynamic programming is employed for family-switch optimization, with state complexity constrained to

2^{n} + 1

. In the last stage of algorithmic framework, we propose a branch-and-bound method to handle family-switch timing, utilizing lower bounds derived from the results of previous stages. Our overall proposed ”branch-and-bound-regulated dynamic programming (B&B-DP)” algorithm excels in solving large-scale scheduling problems, demonstrating superior performance against four benchmark methods across 150 test cases. This algorithmic framework extends the capabilities of single-machine scheduling with family setup times to handle a large number of jobs. In our experiments, we show that the proposed algorithm reduces total tardiness by 10%–25% compared to other methods. This research not only advances the state of the art in single-machine scheduling but also provides a scalable and effective framework for addressing complex production scheduling challenges.

{"title":"Dynamic programming-based exact and heuristic algorithms for single machine scheduling with sequence-dependent setups","authors":"Tengmu Hu , Shih-Hsien Tseng , Theodore T. Allen","doi":"10.1016/j.eswa.2025.126866","DOIUrl":"10.1016/j.eswa.2025.126866","url":null,"abstract":"<div><div>This study presents a novel algorithmic framework and an inventory flow mixed integer programming formulation designed to minimize total tardiness and the number of setups. The approach decomposes the problem into three stages: intra-family scheduling, family sequence optimization, and family-switch timing. We propose a specialized heuristic with <span><math><mrow><mi>O</mi><mrow><mo>(</mo><msup><mrow><mi>n</mi></mrow><mrow><mn>5</mn></mrow></msup><mo>log</mo><mi>n</mi><mo>)</mo></mrow></mrow></math></span> complexity efficiently handles intra-family scheduling and is extended to accommodate subfamily groupings. Dynamic programming is employed for family-switch optimization, with state complexity constrained to <span><math><mrow><msup><mrow><mn>2</mn></mrow><mrow><mi>n</mi></mrow></msup><mo>+</mo><mn>1</mn></mrow></math></span>. In the last stage of algorithmic framework, we propose a branch-and-bound method to handle family-switch timing, utilizing lower bounds derived from the results of previous stages. Our overall proposed ”branch-and-bound-regulated dynamic programming (B&B-DP)” algorithm excels in solving large-scale scheduling problems, demonstrating superior performance against four benchmark methods across 150 test cases. This algorithmic framework extends the capabilities of single-machine scheduling with family setup times to handle a large number of jobs. In our experiments, we show that the proposed algorithm reduces total tardiness by 10%–25% compared to other methods. This research not only advances the state of the art in single-machine scheduling but also provides a scalable and effective framework for addressing complex production scheduling challenges.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126866"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143438082","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

Maximizing data utility while preserving privacy through database fragmentation

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126873

Ali Amiri

Efficiently managing databases that balance data privacy with utility is a critical challenge in today’s data-driven landscape. This study addresses the problem of database fragmentation, which involves dividing a database into smaller fragments, each containing a subset of attributes. The primary objective is to strike a balance between safeguarding the confidentiality of sensitive attribute sets and optimizing the database’s utility. Sensitive attribute sets include combinations of attributes that could disclose private information or identify individuals, such as personal quasi-identifiers, necessitating their separation into distinct fragments to reduce the risk of sensitive data exposure. Conversely, utility attribute sets consist of attributes that enhance data usability and query efficiency. Maximizing utility requires grouping attributes from the same utility set into as few fragments as possible. To effectively solve this complex NP-hard problem, A column generation-based solution leveraging a set partitioning formulation is presented. Experimental evaluations on real and synthetic datasets validate the efficiency of the proposed approach, demonstrating its superiority over the state-of-the-art commercial solver, CPLEX.

{"title":"Maximizing data utility while preserving privacy through database fragmentation","authors":"Ali Amiri","doi":"10.1016/j.eswa.2025.126873","DOIUrl":"10.1016/j.eswa.2025.126873","url":null,"abstract":"<div><div>Efficiently managing databases that balance data privacy with utility is a critical challenge in today’s data-driven landscape. This study addresses the problem of database fragmentation, which involves dividing a database into smaller fragments, each containing a subset of attributes. The primary objective is to strike a balance between safeguarding the confidentiality of sensitive attribute sets and optimizing the database’s utility. Sensitive attribute sets include combinations of attributes that could disclose private information or identify individuals, such as personal quasi-identifiers, necessitating their separation into distinct fragments to reduce the risk of sensitive data exposure. Conversely, utility attribute sets consist of attributes that enhance data usability and query efficiency. Maximizing utility requires grouping attributes from the same utility set into as few fragments as possible. To effectively solve this complex NP-hard problem, A column generation-based solution leveraging a set partitioning formulation is presented. Experimental evaluations on real and synthetic datasets validate the efficiency of the proposed approach, demonstrating its superiority over the state-of-the-art commercial solver, CPLEX.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126873"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143438160","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

Development and validation of a human-machine interface for unmanned aerial vehicle (UAV) control via hand gesture teleoperation

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126828

Fevzi Çakmak Bolat , Mustafa Cem Avci

In this research, a drone-style unmanned aerial vehicle is maneuvered using hand gestures through the creation of a specialized glove design. The analytical formulas pertaining to the drone framework developed during the research were derived, leading to the establishment of a mathematical representation. These formulas were implemented in the Matlab & Simulink environment, and simulations of the system based on this mathematical representation were conducted. Next, to carry out verification tests, a unique device was crafted and set up for the drone, enabling real-time data exchange with the glove. A series of distinct signal sets for the glove were examined to confirm the functionality of the system. After confirming the control mechanism, it was seamlessly incorporated into the electronic hardware framework, leveraging the Arduino Uno microcontroller as the focal point. Within the hand gesture apparatus, an innovative circuit was devised, managed by the Atmega328P microcontroller chip. The primary motivation behind this exploration resides in the desire to establish a user interface for UAV operators that is both seamless and unobtrusive, moving beyond the artificial and cumbersome elements tied to traditional control systems. For this purpose, the research aims to empower users to utilize hand gestures—frequently employed in various everyday scenarios—for piloting activities, thus improving user performance and simplicity of use. The findings of this study highlight the parity between the glove apparatus designed for hand gesture manipulation and the conventional joystick-based system, thereby confirming its effectiveness for multiple applications. Furthermore, a one-handed method was embraced for hand gesture control, with the supplementary aim of offering pilot training opportunities for individuals with upper limb impairments.

{"title":"Development and validation of a human-machine interface for unmanned aerial vehicle (UAV) control via hand gesture teleoperation","authors":"Fevzi Çakmak Bolat , Mustafa Cem Avci","doi":"10.1016/j.eswa.2025.126828","DOIUrl":"10.1016/j.eswa.2025.126828","url":null,"abstract":"<div><div>In this research, a drone-style unmanned aerial vehicle is maneuvered using hand gestures through the creation of a specialized glove design. The analytical formulas pertaining to the drone framework developed during the research were derived, leading to the establishment of a mathematical representation. These formulas were implemented in the Matlab & Simulink environment, and simulations of the system based on this mathematical representation were conducted. Next, to carry out verification tests, a unique device was crafted and set up for the drone, enabling real-time data exchange with the glove. A series of distinct signal sets for the glove were examined to confirm the functionality of the system. After confirming the control mechanism, it was seamlessly incorporated into the electronic hardware framework, leveraging the Arduino Uno microcontroller as the focal point. Within the hand gesture apparatus, an innovative circuit was devised, managed by the Atmega328P microcontroller chip. The primary motivation behind this exploration resides in the desire to establish a user interface for UAV operators that is both seamless and unobtrusive, moving beyond the artificial and cumbersome elements tied to traditional control systems. For this purpose, the research aims to empower users to utilize hand gestures—frequently employed in various everyday scenarios—for piloting activities, thus improving user performance and simplicity of use. The findings of this study highlight the parity between the glove apparatus designed for hand gesture manipulation and the conventional joystick-based system, thereby confirming its effectiveness for multiple applications. Furthermore, a one-handed method was embraced for hand gesture control, with the supplementary aim of offering pilot training opportunities for individuals with upper limb impairments.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126828"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143428993","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

Beyond-Skeleton: Zero-shot Skeleton Action Recognition enhanced by supplementary RGB visual information 超越骨架：通过补充 RGB 视觉信息增强零镜头骨架动作识别能力

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126814

Hongjie Liu , Yingchun Niu , Kun Zeng , Chun Liu , Mengjie Hu , Qing Song

Zero-shot action recognition (ZSAR) recognizes action categories that have not appeared during the training process and has garnered widespread attention due to its potential to save costs in retraining and data annotation. We observed that the existing ZSAR method based on skeleton sequences only uses human posture information in the skeleton sequence, lacks discriminative semantic representation in some similar behavior recognition, and lacks effective interaction between different modalities, resulting in unsatisfactory performance and limited applications of the ZSAR. To solve these problems, we propose a novel method, called Beyond-Skeleton zero-shot Learning (BSZSL), which is used to enhance zero-shot Skeleton Action Recognition. Firstly, a multi-prompt learning strategy is introduced. It utilizes prompt information to guide the model to simultaneously learn complementary semantic information related to behavior categories from both skeleton sequences and RGB information, making the visual feature information more expressive. Specifically, it employs a pre-trained multimodal model to extract prior knowledge related to behaviors from RGB and then guides the skeleton sequence features using this knowledge. This enhances the complementary features of both RGB and skeleton modalities. Secondly, to constrain the mapping relationship of different modal feature information, a Contrastive Clustering (CC) module is designed. This module emphasizes the similarity of features within the same category while increasing the differences in feature mapping between different categories. Finally, evaluating our method on the NTU-60 and NTU-120 datasets with multi-split settings, the result demonstrates that our method achieves state-of-the-art performance in both zero-shot learning (ZSL) and generalized zero-shot learning (GZSL) settings.

{"title":"Beyond-Skeleton: Zero-shot Skeleton Action Recognition enhanced by supplementary RGB visual information","authors":"Hongjie Liu , Yingchun Niu , Kun Zeng , Chun Liu , Mengjie Hu , Qing Song","doi":"10.1016/j.eswa.2025.126814","DOIUrl":"10.1016/j.eswa.2025.126814","url":null,"abstract":"<div><div>Zero-shot action recognition (ZSAR) recognizes action categories that have not appeared during the training process and has garnered widespread attention due to its potential to save costs in retraining and data annotation. We observed that the existing ZSAR method based on skeleton sequences only uses human posture information in the skeleton sequence, lacks discriminative semantic representation in some similar behavior recognition, and lacks effective interaction between different modalities, resulting in unsatisfactory performance and limited applications of the ZSAR. To solve these problems, we propose a novel method, called Beyond-Skeleton zero-shot Learning (BSZSL), which is used to enhance zero-shot Skeleton Action Recognition. Firstly, a multi-prompt learning strategy is introduced. It utilizes prompt information to guide the model to simultaneously learn complementary semantic information related to behavior categories from both skeleton sequences and RGB information, making the visual feature information more expressive. Specifically, it employs a pre-trained multimodal model to extract prior knowledge related to behaviors from RGB and then guides the skeleton sequence features using this knowledge. This enhances the complementary features of both RGB and skeleton modalities. Secondly, to constrain the mapping relationship of different modal feature information, a Contrastive Clustering (CC) module is designed. This module emphasizes the similarity of features within the same category while increasing the differences in feature mapping between different categories. Finally, evaluating our method on the NTU-60 and NTU-120 datasets with multi-split settings, the result demonstrates that our method achieves state-of-the-art performance in both zero-shot learning (ZSL) and generalized zero-shot learning (GZSL) settings.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126814"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143446057","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

Queue length estimation for signal controlling in a connected environment

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126900

Ahmad Abutahoun , Taqwa Alhadidi , Nidhal Saada , Bushra Abutahoun

Queue length at signalized intersection is an important measure that helps design and operate road networks. Queue length information can be shared with road users allowing them to take choose alternative path to avoid delay. In addition, it can be used to determine storage lane length and spacing between two adjacent intersections to avoid grid blockage. This paper focuses on queue length estimation at signalized intersection using Discrete Event Simulation (DES). The developed model using DES is built using Python programming language based on hypothetical data. The model embraces stochastic behavior for both Arrival and departure vehicles at the intersection by following Poisson. Process with exponential distribution of the inter-arrival time. The simulation is done under six different v/c ratios starting from 0.3 to 0.8 with one tenth increment. The results then compared with microsimulation software VISSIM results of the same intersection. The results from DES model and VISSIM model are close to each other with percentage error of 11%. This percentage represent two vehicles at max in the six scenarios. In addition, the developed model does not need calibration nor validation unlike VISSIM. Moreover, the computation time of the developed model found to be significantly faster with 0.2 s while the microsimulation software need 105 s to complete the simulation at maximum speed considering that both models’ simulation time is one hour.

{"title":"Queue length estimation for signal controlling in a connected environment","authors":"Ahmad Abutahoun , Taqwa Alhadidi , Nidhal Saada , Bushra Abutahoun","doi":"10.1016/j.eswa.2025.126900","DOIUrl":"10.1016/j.eswa.2025.126900","url":null,"abstract":"<div><div>Queue length at signalized intersection is an important measure that helps design and operate road networks. Queue length information can be shared with road users allowing them to take choose alternative path to avoid delay. In addition, it can be used to determine storage lane length and spacing between two adjacent intersections to avoid grid blockage. This paper focuses on queue length estimation at signalized intersection using Discrete Event Simulation (DES). The developed model using DES is built using Python programming language based on hypothetical data. The model embraces stochastic behavior for both Arrival and departure vehicles at the intersection by following Poisson. Process with exponential distribution of the inter-arrival time. The simulation is done under six different v/c ratios starting from 0.3 to 0.8 with one tenth increment. The results then compared with microsimulation software VISSIM results of the same intersection. The results from DES model and VISSIM model are close to each other with percentage error of 11%. This percentage represent two vehicles at max in the six scenarios. In addition, the developed model does not need calibration nor validation unlike VISSIM. Moreover, the computation time of the developed model found to be significantly faster with 0.2 s while the microsimulation software need 105 s to complete the simulation at maximum speed considering that both models’ simulation time is one hour.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126900"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464073","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

ODTE—An ensemble of multi-class SVM-based oblique decision trees

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126833

Ricardo Montañana, José A. Gámez, José M. Puerta

We propose ODTE, a new ensemble that uses oblique decision trees as base classifiers. Additionally, we introduce STree, the base algorithm for growing oblique decision trees, which leverages support vector machines to define hyperplanes within the decision nodes. We embed a multiclass strategy (one-vs-one or one-vs-rest) at the decision nodes, allowing the model to directly handle non-binary classification tasks without the need to cluster instances into two groups, as is common in other approaches from the literature. In each decision node, only the best-performing model (SVM)—the one that minimizes an impurity measure for the n-ary classification—is retained, even if the learned SVM addresses a binary classification subtask. An extensive experimental study involving 49 datasets and various state-of-the-art algorithms for oblique decision tree ensembles has been conducted. Our results show that ODTE ranks consistently above its competitors, achieving significant performance gains when hyperparameters are carefully tuned. Moreover, the oblique decision trees learned through STree are more compact than those produced by other algorithms evaluated in our experiments.

{"title":"ODTE—An ensemble of multi-class SVM-based oblique decision trees","authors":"Ricardo Montañana, José A. Gámez, José M. Puerta","doi":"10.1016/j.eswa.2025.126833","DOIUrl":"10.1016/j.eswa.2025.126833","url":null,"abstract":"<div><div>We propose ODTE, a new ensemble that uses oblique decision trees as base classifiers. Additionally, we introduce STree, the base algorithm for growing oblique decision trees, which leverages support vector machines to define hyperplanes within the decision nodes. We embed a multiclass strategy (one-vs-one or one-vs-rest) at the decision nodes, allowing the model to directly handle non-binary classification tasks without the need to cluster instances into two groups, as is common in other approaches from the literature. In each decision node, only the best-performing model (SVM)—the one that minimizes an impurity measure for the n-ary classification—is retained, even if the learned SVM addresses a binary classification subtask. An extensive experimental study involving 49 datasets and various state-of-the-art algorithms for oblique decision tree ensembles has been conducted. Our results show that ODTE ranks consistently above its competitors, achieving significant performance gains when hyperparameters are carefully tuned. Moreover, the oblique decision trees learned through STree are more compact than those produced by other algorithms evaluated in our experiments.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126833"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143429279","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

LRA-GNN: Latent Relation-Aware Graph Neural Network with initial and Dynamic Residual for facial age estimation

IF 7.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Expert Systems with Applications

Pub Date : 2025-02-15 DOI: 10.1016/j.eswa.2025.126819

Yiping Zhang , Yuntao Shou , Wei Ai , Tao Meng , Keqin Li

Face information is mainly concentrated among facial key points, and frontier research has begun to use graph neural networks to segment faces into patches as nodes to model complex face representations. However, these methods construct node-to-node relations based on similarity thresholds, so there is a problem that some latent relations are missing. These latent relations are crucial for deep semantic representation of face aging. In this novel, we propose a new Latent Relation-Aware Graph Neural Network with Initial and Dynamic Residual (LRA-GNN) to achieve robust and comprehensive facial representation. Specifically, we first construct an initial graph utilizing facial key points as prior knowledge, and then a random walk strategy is employed to the initial graph for obtaining the global structure, both of which together guide the subsequent effective exploration and comprehensive representation. Then LRA-GNN leverages the multi-attention mechanism to capture the latent relations and generates a set of fully connected graphs containing rich facial information and complete structure based on the aforementioned guidance. To avoid over-smoothing issues for deep feature extraction on the fully connected graphs, the deep residual graph convolutional networks are carefully designed, which fuse adaptive initial residuals and dynamic developmental residuals to ensure the consistency and diversity of information. Finally, to improve the estimation accuracy and generalization ability, progressive reinforcement learning is proposed to optimize the ensemble classification regressor. Our proposed framework surpasses the state-of-the-art baselines on several age estimation benchmarks, demonstrating its strength and effectiveness.

{"title":"LRA-GNN: Latent Relation-Aware Graph Neural Network with initial and Dynamic Residual for facial age estimation","authors":"Yiping Zhang , Yuntao Shou , Wei Ai , Tao Meng , Keqin Li","doi":"10.1016/j.eswa.2025.126819","DOIUrl":"10.1016/j.eswa.2025.126819","url":null,"abstract":"<div><div>Face information is mainly concentrated among facial key points, and frontier research has begun to use graph neural networks to segment faces into patches as nodes to model complex face representations. However, these methods construct node-to-node relations based on similarity thresholds, so there is a problem that some latent relations are missing. These latent relations are crucial for deep semantic representation of face aging. In this novel, we propose a new Latent Relation-Aware Graph Neural Network with Initial and Dynamic Residual (LRA-GNN) to achieve robust and comprehensive facial representation. Specifically, we first construct an initial graph utilizing facial key points as prior knowledge, and then a random walk strategy is employed to the initial graph for obtaining the global structure, both of which together guide the subsequent effective exploration and comprehensive representation. Then LRA-GNN leverages the multi-attention mechanism to capture the latent relations and generates a set of fully connected graphs containing rich facial information and complete structure based on the aforementioned guidance. To avoid over-smoothing issues for deep feature extraction on the fully connected graphs, the deep residual graph convolutional networks are carefully designed, which fuse adaptive initial residuals and dynamic developmental residuals to ensure the consistency and diversity of information. Finally, to improve the estimation accuracy and generalization ability, progressive reinforcement learning is proposed to optimize the ensemble classification regressor. Our proposed framework surpasses the state-of-the-art baselines on several age estimation benchmarks, demonstrating its strength and effectiveness.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126819"},"PeriodicalIF":7.5,"publicationDate":"2025-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464072","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