Expert Systems with Applications最新文献_第6页

Multichannel feature fusion network-based technique for heart sound signal classification and recognition

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

Expert Systems with Applications

Pub Date : 2025-02-19 DOI: 10.1016/j.eswa.2025.126839

Weihua Xiong, Guan Zhang, Dongming Yan, Lixian Cao, Xiaotong Huang, Du Li

Heart sound signals are widely used in medical applications for disease prevention, initial diagnosis, and long-term monitoring of patient conditions. Accurate processing and analysis of heart sound signals allow doctors to better understand the patient’s condition and formulate more appropriate prevention and treatment plans. However, the physician’s recognition of heart sound signals from time series cannot exclude interference from subjective factors when processing such high-dimensional data, resulting in inaccurate recognition results. Additionally, with traditional machine learning methods, further improvement is difficult to achieve, and existing neural network algorithms do not effectively utilize the long-term contextual relationship of time series signals. To address these problems, this study constructed an end-to-end neural network sequence labeling algorithm based on the physical information of heart sound signals and embedded a saliency attentive model network (SAM-Net) module to reduce interference from redundant information. The results of the labeling algorithm were used to design a multichannel feature fusion network for heart sound signals, incorporating a squeeze excitation network (SE-Net) module to accelerate the extraction of target features in different channels, which is different from the traditional classify, recognize, detect, and analyze approach. The proposed method improved robustness and adaptability of classification and recognition of heart sound signals, performing well on the selected dataset, thereby obtaining the highest recognition accuracy of 97.23 % and F₁ score of 97.08 %. These results are significantly better than previous classification methods by other researchers. This work provides a clinical informatics tool to assist clinician with early detection of abnormal heart conditions.

{"title":"Multichannel feature fusion network-based technique for heart sound signal classification and recognition","authors":"Weihua Xiong, Guan Zhang, Dongming Yan, Lixian Cao, Xiaotong Huang, Du Li","doi":"10.1016/j.eswa.2025.126839","DOIUrl":"10.1016/j.eswa.2025.126839","url":null,"abstract":"<div><div>Heart sound signals are widely used in medical applications for disease prevention, initial diagnosis, and long-term monitoring of patient conditions. Accurate processing and analysis of heart sound signals allow doctors to better understand the patient’s condition and formulate more appropriate prevention and treatment plans. However, the physician’s recognition of heart sound signals from time series cannot exclude interference from subjective factors when processing such high-dimensional data, resulting in inaccurate recognition results. Additionally, with traditional machine learning methods, further improvement is difficult to achieve, and existing neural network algorithms do not effectively utilize the long-term contextual relationship of time series signals. To address these problems, this study constructed an end-to-end neural network sequence labeling algorithm based on the physical information of heart sound signals and embedded a saliency attentive model network (SAM-Net) module to reduce interference from redundant information. The results of the labeling algorithm were used to design a multichannel feature fusion network for heart sound signals, incorporating a squeeze excitation network (SE-Net) module to accelerate the extraction of target features in different channels, which is different from the traditional classify, recognize, detect, and analyze approach. The proposed method improved robustness and adaptability of classification and recognition of heart sound signals, performing well on the selected dataset, thereby obtaining the highest recognition accuracy of 97.23 % and F<sub>1</sub> score of 97.08 %. These results are significantly better than previous classification methods by other researchers. This work provides a clinical informatics tool to assist clinician with early detection of abnormal heart conditions.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126839"},"PeriodicalIF":7.5,"publicationDate":"2025-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464060","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

Gaseous fuel supply chain configuration selection: A life cycle thinking-based decision support framework

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

Expert Systems with Applications

Pub Date : 2025-02-19 DOI: 10.1016/j.eswa.2025.126944

Ravihari Kotagodahetti , Kasun Hewage , Ezzeddin Bakhtavar , Rehan Sadiq

This paper analyzes the impact of stakeholder priorities and causal relationships between decision criteria when determining the most suitable renewable natural gas (RNG) and hydrogen gaseous fuels production path. The criteria indicators were defined to reflect the life cycle of environmental desirability and economic feasibility. The study introduces a framework-based approach using fuzzy multi-objective optimization by ratio analysis (FMOORA) with an integrated causality and importance weighting system. The core novelty of this study is the novel criterion weighting system that simultaneously considers the impacts of the internal and external causal relationships and the criteria importance arising from stakeholder priorities to derive a more realistic weighting scheme. Weighting schemes using fuzzy cognitive maps and the best-worst method were provided to consider the decision-maker’s priorities in highlighting each causality or importance weighting concept. Accordingly, the highest importance values of 0.152 and 0.267 were obtained for levelized cost of energy with fuzzy cognitive maps and the best-worst method, respectively. The profitability index scored the least importance values in both methods (fuzzy cognitive map – 0.071 and best-worst method – 0.038). Accordingly, RNG from livestock with pressure swing adsorption was ranked first under all weighting schemes. Hydrogen from steam methane reforming achieved the best spot under hydrogen production routes, with ranks varying from 8 to 12 when both RNG and hydrogen production scenarios are considered. The findings indicate that the proposed framework is useful for conducting preliminary feasibility assessments of gaseous fuel investment strategies. Additionally, the integrated weighting system can be considered in conjunction with other multi-criteria decision-making techniques to make more reliable decisions in different problems.

{"title":"Gaseous fuel supply chain configuration selection: A life cycle thinking-based decision support framework","authors":"Ravihari Kotagodahetti , Kasun Hewage , Ezzeddin Bakhtavar , Rehan Sadiq","doi":"10.1016/j.eswa.2025.126944","DOIUrl":"10.1016/j.eswa.2025.126944","url":null,"abstract":"<div><div>This paper analyzes the impact of stakeholder priorities and causal relationships between decision criteria when determining the most suitable renewable natural gas (RNG) and hydrogen gaseous fuels production path. The criteria indicators were defined to reflect the life cycle of environmental desirability and economic feasibility. The study introduces a framework-based approach using fuzzy multi-objective optimization by ratio analysis (FMOORA) with an integrated causality and importance weighting system. The core novelty of this study is the novel criterion weighting system that simultaneously considers the impacts of the internal and external causal relationships and the criteria importance arising from stakeholder priorities to derive a more realistic weighting scheme. Weighting schemes using fuzzy cognitive maps and the best-worst method were provided to consider the decision-maker’s priorities in highlighting each causality or importance weighting concept. Accordingly, the highest importance values of 0.152 and 0.267 were obtained for levelized cost of energy with fuzzy cognitive maps and the best-worst method, respectively. The profitability index scored the least importance values in both methods (fuzzy cognitive map – 0.071 and best-worst method – 0.038). Accordingly, RNG from livestock with pressure swing adsorption was ranked first under all weighting schemes. Hydrogen from steam methane reforming achieved the best spot under hydrogen production routes, with ranks varying from 8 to 12 when both RNG and hydrogen production scenarios are considered. The findings indicate that the proposed framework is useful for conducting preliminary feasibility assessments of gaseous fuel investment strategies. Additionally, the integrated weighting system can be considered in conjunction with other multi-criteria decision-making techniques to make more reliable decisions in different problems.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126944"},"PeriodicalIF":7.5,"publicationDate":"2025-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143438085","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

P-TTAN: A novel neural network optimized for thermal feature perception and representation in 3D temperature predictions

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

Expert Systems with Applications

Pub Date : 2025-02-19 DOI: 10.1016/j.eswa.2025.126964

Jincheng Chen , Feiding Zhu , Yuge Han , Dengfeng Ren , Qunqing Lin

Thermal properties of armored vehicles are significant for both offensive and defensive strategies in modern warfare and surveillance. Existing methods struggle with the computational complexity, weak generalization capabilities and lack the real-time capability necessary for dynamic battlefield conditions. This paper presents a novel Parametric Transient Thermophysical Attention Network (P-TTAN), which can predict three-dimensional (3D) transient temperature fields of armored vehicles. Utilizing 3D shape data, material parameters, and environmental meteorological conditions, P-TTAN demonstrates significant advancements in predictive accuracy and computational efficiency. Inspired by PointNet structure, our network integrates shape and thermal features effectively, enhancing its ability to process complex 3D point cloud data. By designing a newly thermal feature extraction module, P-TTAN can recognize both transient and invariant parameters via an improved Transformer. Moreover, P-TTAN incorporates heat conduction differential equations via a simplified method, enhancing the interpretability and reliability of its predictions while reducing computational costs and improving its applicability. The results on the test dataset indicate that the 3D temperature field predictions made by P-TTAN are markedly superior to other state-of-the-art networks, achieving a Mean Absolute Error of just 0.5687 K. This capability marks it as a highly effective tool for real-time thermal analysis in both military and civilian contexts.

{"title":"P-TTAN: A novel neural network optimized for thermal feature perception and representation in 3D temperature predictions","authors":"Jincheng Chen , Feiding Zhu , Yuge Han , Dengfeng Ren , Qunqing Lin","doi":"10.1016/j.eswa.2025.126964","DOIUrl":"10.1016/j.eswa.2025.126964","url":null,"abstract":"<div><div>Thermal properties of armored vehicles are significant for both offensive and defensive strategies in modern warfare and surveillance. Existing methods struggle with the computational complexity, weak generalization capabilities and lack the real-time capability necessary for dynamic battlefield conditions. This paper presents a novel Parametric Transient Thermophysical Attention Network (P-TTAN), which can predict three-dimensional (3D) transient temperature fields of armored vehicles. Utilizing 3D shape data, material parameters, and environmental meteorological conditions, P-TTAN demonstrates significant advancements in predictive accuracy and computational efficiency. Inspired by PointNet structure, our network integrates shape and thermal features effectively, enhancing its ability to process complex 3D point cloud data. By designing a newly thermal feature extraction module, P-TTAN can recognize both transient and invariant parameters via an improved Transformer. Moreover, P-TTAN incorporates heat conduction differential equations via a simplified method, enhancing the interpretability and reliability of its predictions while reducing computational costs and improving its applicability. The results on the test dataset indicate that the 3D temperature field predictions made by P-TTAN are markedly superior to other state-of-the-art networks, achieving a Mean Absolute Error of just 0.5687 K. This capability marks it as a highly effective tool for real-time thermal analysis in both military and civilian contexts.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"274 ","pages":"Article 126964"},"PeriodicalIF":7.5,"publicationDate":"2025-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143488653","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

Multi-task multi-view and iterative error-correcting random forest for acute toxicity prediction

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

Expert Systems with Applications

Pub Date : 2025-02-18 DOI: 10.1016/j.eswa.2025.126972

Jie Gao , Lianlian Wu , Guangyi Lin , Jiayu Zou , Bowei Yan , Kunhong Liu , Song He , Xiaochen Bo

Unexpected toxicity poses a significant impediment to successful entry of drug candidates into the market. For drug toxicity evaluation, deep learning techniques have exhibited robust competitiveness compared to costly and ethically challenging animal studies. However, recent modeling primarily relies on deep neural networks with limited attention to species with small sample sizes and lacking interpretability. In response to these concerns, we proposed an innovative variant algorithm based on random forest, termed the Multi-Task Iterative Error-Correcting Random Forest (MTIEC-RF), to predict multi-species acute toxicity using both single-view and multi-view approaches. In the single-view context, MTIEC-RF utilized the multi-task random forest as its backbone structure. Through iterative processes, it generated a set of error-correcting decision trees based on challenging samples associated with diverse toxicity endpoints. In the multi-view scenarios, we integrated reliable multi-view data, a consensus framework, and endpoint representations. The final MTIEC-RF model is with better generalization ability on medium- and small-sized toxicity endpoints, and significantly outperformed state-of-the-art deep learning techniques by achieving 14% performance improvement on the average R² of 59 datasets. Additionally, MTIEC-RF identified pivotal descriptors contributing to acute toxicity prediction by quantifying feature importance, number of hydrogen atoms emerged as a significant influential factor in the prediction process.

{"title":"Multi-task multi-view and iterative error-correcting random forest for acute toxicity prediction","authors":"Jie Gao , Lianlian Wu , Guangyi Lin , Jiayu Zou , Bowei Yan , Kunhong Liu , Song He , Xiaochen Bo","doi":"10.1016/j.eswa.2025.126972","DOIUrl":"10.1016/j.eswa.2025.126972","url":null,"abstract":"<div><div>Unexpected toxicity poses a significant impediment to successful entry of drug candidates into the market. For drug toxicity evaluation, deep learning techniques have exhibited robust competitiveness compared to costly and ethically challenging animal studies. However, recent modeling primarily relies on deep neural networks with limited attention to species with small sample sizes and lacking interpretability. In response to these concerns, we proposed an innovative variant algorithm based on random forest, termed the Multi-Task Iterative Error-Correcting Random Forest (MTIEC-RF), to predict multi-species acute toxicity using both single-view and multi-view approaches. In the single-view context, MTIEC-RF utilized the multi-task random forest as its backbone structure. Through iterative processes, it generated a set of error-correcting decision trees based on challenging samples associated with diverse toxicity endpoints. In the multi-view scenarios, we integrated reliable multi-view data, a consensus framework, and endpoint representations. The final MTIEC-RF model is with better generalization ability on medium- and small-sized toxicity endpoints, and significantly outperformed state-of-the-art deep learning techniques by achieving 14% performance improvement on the average <em>R<sup>2</sup></em> of 59 datasets. Additionally, MTIEC-RF identified pivotal descriptors contributing to acute toxicity prediction by quantifying feature importance, number of hydrogen atoms emerged as a significant influential factor in the prediction process.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"274 ","pages":"Article 126972"},"PeriodicalIF":7.5,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143474359","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

Hist2Vec: A histogram and kernel-based embedding method for molecular sequence analysis

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

Expert Systems with Applications

Pub Date : 2025-02-18 DOI: 10.1016/j.eswa.2025.126859

Sarwan Ali , Tamkanat E. Ali , Haris Mansoor , Prakash Chourasia , Murray Patterson

Due to the huge surge in genomic data, there is an increasing need for better and more efficient molecular sequence classification techniques. There has been plenty of work proposed by researchers using machine learning models for promising classification results. However, they face few limitations in capturing hierarchical structures and relationships in the molecular sequences. To overcome such limitations, we propose Hist2Vec, a novel kernel-based technique for embedding generation that captures the sequence similarities by constructing histogram-based kernel matrices and Gaussian kernel functions. By building histogram-based representations from the distinct

k

-mers and minimizers found in each sequence, Hist2Vec is able to identify similarities between sequences. The sequence information is preserved by converting these representations to higher dimensional feature spaces using Gaussian Kernel functions. Then we apply kernel Principal Component Analysis to obtain the final embedding for the molecular sequences. These embeddings are then used as input to classical machine learning models for supervised analysis. We also establish the theoretical properties of Hist2Vec, ensuring the validity and effectiveness of the method. The experimental evaluation of our method shows that Hist2Vec outperforms all other state-of-the-art methods demonstrating high accuracy of

> 76 %

for the Human DNA dataset,

> 83 %

for the Coronavirus Host dataset, and high precision in the case of t-Cell dataset.

{"title":"Hist2Vec: A histogram and kernel-based embedding method for molecular sequence analysis","authors":"Sarwan Ali , Tamkanat E. Ali , Haris Mansoor , Prakash Chourasia , Murray Patterson","doi":"10.1016/j.eswa.2025.126859","DOIUrl":"10.1016/j.eswa.2025.126859","url":null,"abstract":"<div><div>Due to the huge surge in genomic data, there is an increasing need for better and more efficient molecular sequence classification techniques. There has been plenty of work proposed by researchers using machine learning models for promising classification results. However, they face few limitations in capturing hierarchical structures and relationships in the molecular sequences. To overcome such limitations, we propose Hist2Vec, a novel kernel-based technique for embedding generation that captures the sequence similarities by constructing histogram-based kernel matrices and Gaussian kernel functions. By building histogram-based representations from the distinct <span><math><mi>k</mi></math></span>-mers and minimizers found in each sequence, Hist2Vec is able to identify similarities between sequences. The sequence information is preserved by converting these representations to higher dimensional feature spaces using Gaussian Kernel functions. Then we apply kernel Principal Component Analysis to obtain the final embedding for the molecular sequences. These embeddings are then used as input to classical machine learning models for supervised analysis. We also establish the theoretical properties of Hist2Vec, ensuring the validity and effectiveness of the method. The experimental evaluation of our method shows that Hist2Vec outperforms all other state-of-the-art methods demonstrating high accuracy of <span><math><mrow><mo>></mo><mn>76</mn><mtext>%</mtext></mrow></math></span> for the Human DNA dataset, <span><math><mrow><mo>></mo><mn>83</mn><mtext>%</mtext></mrow></math></span> for the Coronavirus Host dataset, and high precision in the case of t-Cell dataset.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126859"},"PeriodicalIF":7.5,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143453665","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

MTLSER: Multi-task learning enhanced speech emotion recognition with pre-trained acoustic model

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

Expert Systems with Applications

Pub Date : 2025-02-18 DOI: 10.1016/j.eswa.2025.126855

Zengzhao Chen , Chuan Liu , Zhifeng Wang , Chuanxu Zhao , Mengting Lin , Qiuyu Zheng

This study proposes a novel Speech Emotion Recognition (SER) approach employing a Multi-Task Learning framework (MTLSER), designed to boost recognition accuracy by training multiple related tasks simultaneously and sharing information via a joint loss function. This framework integrates SER as the primary task, with Automatic Speech Recognition (ASR) and speaker identification serving as auxiliary tasks. Feature extraction is conducted using the pre-trained wav2vec2.0 model, which acts as a shared layer within our multi-task learning (MTL) framework. Extracted features are then processed in parallel by the three tasks. The contributions of auxiliary tasks are adjusted through hyperparameters, and their loss functions are amalgamated into a singular joint loss function for effective backpropagation. This optimization refines the model’s internal parameters. Our method’s efficacy is tested during the inference stage, where the model concurrently outputs the emotion, textual content, and speaker identity from the input audio. We conducted ablation studies and a sensitivity analysis on the hyperparameters to determine the optimal settings for emotion recognition. The performance of our proposed MTLSER model is evaluated using the public IEMOCAP dataset. Results from extensive testing show a significant improvement over traditional methods, achieving a Weighted Accuracy (WA) of 82.63% and an Unweighted Accuracy (UA) of 82.19%. These findings affirm the effectiveness and robustness of our approach. Our code is publicly available at https://github.com/CCNU-nercel-lc/MTL-SER.

{"title":"MTLSER: Multi-task learning enhanced speech emotion recognition with pre-trained acoustic model","authors":"Zengzhao Chen , Chuan Liu , Zhifeng Wang , Chuanxu Zhao , Mengting Lin , Qiuyu Zheng","doi":"10.1016/j.eswa.2025.126855","DOIUrl":"10.1016/j.eswa.2025.126855","url":null,"abstract":"<div><div>This study proposes a novel Speech Emotion Recognition (SER) approach employing a Multi-Task Learning framework (MTLSER), designed to boost recognition accuracy by training multiple related tasks simultaneously and sharing information via a joint loss function. This framework integrates SER as the primary task, with Automatic Speech Recognition (ASR) and speaker identification serving as auxiliary tasks. Feature extraction is conducted using the pre-trained wav2vec2.0 model, which acts as a shared layer within our multi-task learning (MTL) framework. Extracted features are then processed in parallel by the three tasks. The contributions of auxiliary tasks are adjusted through hyperparameters, and their loss functions are amalgamated into a singular joint loss function for effective backpropagation. This optimization refines the model’s internal parameters. Our method’s efficacy is tested during the inference stage, where the model concurrently outputs the emotion, textual content, and speaker identity from the input audio. We conducted ablation studies and a sensitivity analysis on the hyperparameters to determine the optimal settings for emotion recognition. The performance of our proposed MTLSER model is evaluated using the public IEMOCAP dataset. Results from extensive testing show a significant improvement over traditional methods, achieving a Weighted Accuracy (WA) of 82.63% and an Unweighted Accuracy (UA) of 82.19%. These findings affirm the effectiveness and robustness of our approach. Our code is publicly available at <span><span>https://github.com/CCNU-nercel-lc/MTL-SER</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126855"},"PeriodicalIF":7.5,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143438159","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

KanDMVC: KAN be used for deep multi-view clustering?

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

Expert Systems with Applications

Pub Date : 2025-02-18 DOI: 10.1016/j.eswa.2025.126868

Yifan Zhang , Yong Wang , Guifu Lu , Cuiyun Gao , Lin Cui , Zizhuang Ma

Deep multi-view clustering has garnered increasing attention for its ability to explore common semantics from multiple views for clustering tasks using deep learning techniques. However, existing research faces two challenges. (1) Coding dependency challenge: Current coding frameworks are composed of linear layers and the same activation function stacked together, resulting in an extreme dependence on the choice of activation function for its ability to capture features. (2) Data discrepancy challenge: Different view data exhibit certain variability due to originating from different data sources. These variabilities can affect the results of feature fusion, thereby reducing clustering performance. To address these challenges, this paper proposes a view comparison fusion framework based on the Kolmogorov–Arnold Network for deep multi-view clustering (KanDMVC). First, we design a new adaptive activation function coding framework based on KAN (KANencoder), which adaptively learns the matching activation function during training on different datasets to reducing the challenge of coding dependency. Additionally, to fully consider the differences and similarities between view features, we design a contrast fusion module (Feature fusion) that takes into account feature variability and similarity, addressing the data discrepancy challenge to learn a more comprehensive feature representation. Finally, extensive experiments on multiple datasets demonstrate that our proposed method outperforms state-of-the-art deep multi-view clustering algorithms. The code for this article is published on Github at https://github.com/snothingtosay/KanDMVC.git.

{"title":"KanDMVC: KAN be used for deep multi-view clustering?","authors":"Yifan Zhang , Yong Wang , Guifu Lu , Cuiyun Gao , Lin Cui , Zizhuang Ma","doi":"10.1016/j.eswa.2025.126868","DOIUrl":"10.1016/j.eswa.2025.126868","url":null,"abstract":"<div><div>Deep multi-view clustering has garnered increasing attention for its ability to explore common semantics from multiple views for clustering tasks using deep learning techniques. However, existing research faces two challenges. (1) Coding dependency challenge: Current coding frameworks are composed of linear layers and the same activation function stacked together, resulting in an extreme dependence on the choice of activation function for its ability to capture features. (2) Data discrepancy challenge: Different view data exhibit certain variability due to originating from different data sources. These variabilities can affect the results of feature fusion, thereby reducing clustering performance. To address these challenges, this paper proposes a view comparison fusion framework based on the Kolmogorov–Arnold Network for deep multi-view clustering (KanDMVC). First, we design a new adaptive activation function coding framework based on KAN (KANencoder), which adaptively learns the matching activation function during training on different datasets to reducing the challenge of coding dependency. Additionally, to fully consider the differences and similarities between view features, we design a contrast fusion module (Feature fusion) that takes into account feature variability and similarity, addressing the data discrepancy challenge to learn a more comprehensive feature representation. Finally, extensive experiments on multiple datasets demonstrate that our proposed method outperforms state-of-the-art deep multi-view clustering algorithms. The code for this article is published on Github at <span><span>https://github.com/snothingtosay/KanDMVC.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126868"},"PeriodicalIF":7.5,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143453660","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

XAI in hindsight: Shapley values for explaining prediction accuracy

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

Expert Systems with Applications

Pub Date : 2025-02-18 DOI: 10.1016/j.eswa.2025.126845

Andreas Brandsæter , Ingrid K. Glad

Predicting the outcome of AI-models is inherently difficult, and understanding and trusting the models and decisions based on them are challenging. To help us, various explainable artificial intelligence (XAI) methods have been developed. Sometimes, explanations are requested in hindsight, for example after an accident has occurred due to unexpected or erroneous model outcomes. In such situations, our focus is on answering why the model failed to produce an accurate prediction. But since XAI-methods are typically made without knowledge of the true outcome values, the explanations concern the prediction and not the prediction error. In this paper, we change perspective and assume that the true values are known and propose an explanation method that quantifies how different subsets/clusters of the training data impact how the predicted values deviate from the true values, the so-called residuals. In this way, the proposed method lets us explain the accuracy of individual predictions, in hindsight. By focusing on explanations in hindsight, rather than the predictions per se, the proposed method offers a novel perspective to the field of XAI. The method is demonstrated and evaluated using both synthetic and real-world data. To objectively evaluate the method we propose, we utilize the explanations we generate to tailor a training data acquisition strategy and show how this leads to improved prediction performance. The proposed method is fully generic, and applicable to any industry. In the presentation offered here, most examples are related to the maritime industry.

{"title":"XAI in hindsight: Shapley values for explaining prediction accuracy","authors":"Andreas Brandsæter , Ingrid K. Glad","doi":"10.1016/j.eswa.2025.126845","DOIUrl":"10.1016/j.eswa.2025.126845","url":null,"abstract":"<div><div>Predicting the outcome of AI-models is inherently difficult, and understanding and trusting the models and decisions based on them are challenging. To help us, various explainable artificial intelligence (XAI) methods have been developed. Sometimes, explanations are requested in hindsight, for example after an accident has occurred due to unexpected or erroneous model outcomes. In such situations, our focus is on answering why the model failed to produce an accurate prediction. But since XAI-methods are typically made without knowledge of the true outcome values, the explanations concern the prediction and not the prediction error. In this paper, we change perspective and assume that the true values are known and propose an explanation method that quantifies how different subsets/clusters of the training data impact how the predicted values deviate from the true values, the so-called residuals. In this way, the proposed method lets us explain the accuracy of individual predictions, in hindsight. By focusing on explanations in hindsight, rather than the predictions per se, the proposed method offers a novel perspective to the field of XAI. The method is demonstrated and evaluated using both synthetic and real-world data. To objectively evaluate the method we propose, we utilize the explanations we generate to tailor a training data acquisition strategy and show how this leads to improved prediction performance. The proposed method is fully generic, and applicable to any industry. In the presentation offered here, most examples are related to the maritime industry.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126845"},"PeriodicalIF":7.5,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143454408","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Domain adaptation extreme learning machines for regression and their application in precise modeling of cutting forces under small sample conditions

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

Expert Systems with Applications

Pub Date : 2025-02-18 DOI: 10.1016/j.eswa.2025.126967

Shaonan Zhang , Liangshan Xiong

Existing transfer learning algorithms for cutting forces modeling do not consider the impact of conditional shift on prediction results, leading to principle errors. When the sample size of the target domain dataset is small, these principle errors become more pronounced, resulting in larger prediction errors that cannot be ignored. To address this issue, we propose two transfer learning algorithms that consider the conditional shift, specifically designed for regression tasks such as cutting forces modeling: the regression domain adaptation extreme learning machine algorithms RDAELM-CEOD-A and RDAELM-CEOD-B. These two algorithms, compared to those designed for classification tasks, eliminate conditional shift by minimizing the conditional embedding operator discrepancy instead of maximum mean discrepancy, which makes them more suitable for regression tasks. With the large sample theoretical dataset of cutting forces (calculated from the unequal division shear zone model) as the source domain dataset and the small sample experimental dataset (obtained from cutting experiments) as the target domain dataset, we employ the proposed algorithms to develop prediction models of cutting forces in orthogonal cutting of 6061-T6 aluminum alloy and 42CrMo4 steel. Simulation of the constructed model shows that the prediction errors decrease as the sample size of the target domain dataset increases and are significantly smaller than those of similar models built using other domain adaptation extreme learning machine algorithms. The research findings can be extended to the precise transfer learning modeling of other continuously varying physical quantities under small sample conditions.

{"title":"Domain adaptation extreme learning machines for regression and their application in precise modeling of cutting forces under small sample conditions","authors":"Shaonan Zhang , Liangshan Xiong","doi":"10.1016/j.eswa.2025.126967","DOIUrl":"10.1016/j.eswa.2025.126967","url":null,"abstract":"<div><div>Existing transfer learning algorithms for cutting forces modeling do not consider the impact of conditional shift on prediction results, leading to principle errors. When the sample size of the target domain dataset is small, these principle errors become more pronounced, resulting in larger prediction errors that cannot be ignored. To address this issue, we propose two transfer learning algorithms that consider the conditional shift, specifically designed for regression tasks such as cutting forces modeling: the regression domain adaptation extreme learning machine algorithms RDAELM-CEOD-A and RDAELM-CEOD-B. These two algorithms, compared to those designed for classification tasks, eliminate conditional shift by minimizing the conditional embedding operator discrepancy instead of maximum mean discrepancy, which makes them more suitable for regression tasks. With the large sample theoretical dataset of cutting forces (calculated from the unequal division shear zone model) as the source domain dataset and the small sample experimental dataset (obtained from cutting experiments) as the target domain dataset, we employ the proposed algorithms to develop prediction models of cutting forces in orthogonal cutting of 6061-T6 aluminum alloy and 42CrMo4 steel. Simulation of the constructed model shows that the prediction errors decrease as the sample size of the target domain dataset increases and are significantly smaller than those of similar models built using other domain adaptation extreme learning machine algorithms. The research findings can be extended to the precise transfer learning modeling of other continuously varying physical quantities under small sample conditions.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"274 ","pages":"Article 126967"},"PeriodicalIF":7.5,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143479889","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

FedHNR: Federated hierarchical resilient learning for echocardiogram segmentation with annotation noise

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

Expert Systems with Applications

Pub Date : 2025-02-18 DOI: 10.1016/j.eswa.2025.126841

Yiwen Wang , Wanli Ding , Weiyuan Lin , Tao Tan , Zhifan Gao

Echocardiogram segmentation based on federated learning plays a critical role in enhancing diagnostic accuracy and efficiency. However, challenges such as inter-client annotation noise, client heterogeneity, and limited expert annotations hinder the echocardiogram segmentation based on federated learning. To address these challenges, we propose FedHNR, a federated hierarchical noise-resilient method that identifies and leverages annotation noise across global and local hierarchies. At the global-hierarchy, expert samples fine-tune the global model through a novel weight noise decoupling approach, reducing overfitting while preserving aggregated client knowledge. At the local-hierarchy, FedHNR employs region-level noise assessment and sample-level noise calibration to refine annotations using pseudo-clean labels derived from the global model. These hierarchies together mitigate the negativeness of noise and enhance the model robustness to noise. Extensive experiments on 95,469 echocardiogram frames across public and private datasets demonstrate that FedHNR outperforms ten state-of-the-art methods, showcasing its robustness in both traditional federated learning and real-world scenarios.

基于联合学习的超声心动图分割在提高诊断准确性和效率方面起着至关重要的作用。然而，客户端间注释噪声、客户端异质性和有限的专家注释等挑战阻碍了基于联合学习的超声心动图分割。为了应对这些挑战，我们提出了一种联合分层抗噪方法 FedHNR，它能识别并利用全局和局部分层的注释噪声。在全局层次结构中，专家样本通过一种新颖的权重噪声解耦方法对全局模型进行微调，从而减少过拟合，同时保留聚合的客户知识。在局部层次上，FedHNR 采用区域级噪声评估和样本级噪声校准，利用从全局模型中提取的伪清洁标签完善注释。这些层次结构共同减轻了噪声的负面影响，增强了模型对噪声的鲁棒性。在公共和私有数据集的 95,469 个超声心动图帧上进行的广泛实验表明，FedHNR 优于十种最先进的方法，展示了它在传统联合学习和现实世界场景中的鲁棒性。

{"title":"FedHNR: Federated hierarchical resilient learning for echocardiogram segmentation with annotation noise","authors":"Yiwen Wang , Wanli Ding , Weiyuan Lin , Tao Tan , Zhifan Gao","doi":"10.1016/j.eswa.2025.126841","DOIUrl":"10.1016/j.eswa.2025.126841","url":null,"abstract":"<div><div>Echocardiogram segmentation based on federated learning plays a critical role in enhancing diagnostic accuracy and efficiency. However, challenges such as inter-client annotation noise, client heterogeneity, and limited expert annotations hinder the echocardiogram segmentation based on federated learning. To address these challenges, we propose FedHNR, a federated hierarchical noise-resilient method that identifies and leverages annotation noise across global and local hierarchies. At the global-hierarchy, expert samples fine-tune the global model through a novel weight noise decoupling approach, reducing overfitting while preserving aggregated client knowledge. At the local-hierarchy, FedHNR employs region-level noise assessment and sample-level noise calibration to refine annotations using pseudo-clean labels derived from the global model. These hierarchies together mitigate the negativeness of noise and enhance the model robustness to noise. Extensive experiments on 95,469 echocardiogram frames across public and private datasets demonstrate that FedHNR outperforms ten state-of-the-art methods, showcasing its robustness in both traditional federated learning and real-world scenarios.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"273 ","pages":"Article 126841"},"PeriodicalIF":7.5,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143454405","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