Pub Date : 2025-03-03DOI: 10.26599/TST.2024.9010075
Yasir Glani;Luo Ping;Syed Asad Shah;Lin Ke
The rapid evolution of software development has accentuated the deficiencies of prevailing code clone detection techniques. As modern applications become more complex, traditional cloning tools often struggle to detect general and large-gap clones that undergo regular modification. Such challenges pose threats to software integrity, emphasizing the critical need for improved code cloning techniques. Observing the prevailing gap, we propose an innovative code clone dive (CCDive) code cloning technique, which is designed to detect an extensive range of clones, from direct clones to the often challenging large-gap clones, thoroughly covering different categories, such as very strongly Type-III, strongly Type-III, and moderate Type-III clones. In CCDive, the fusion of a level-by-level abstraction and an innovative similarity matching algorithm ensures the recognition of clones even when nearly half the original code in the chunk has been modified. Furthermore, by integrating the Smith-Waterman local sequence alignment, the capability of CCDive to spot exact code transformation locations can be enhanced. In a comprehensive evaluation, CCDive was compared with well-known code cloning techniques. The efficacy of CCDive was measured using precision, recall, F1-score, accuracy, and efficiency. CCDive consistently surpassed other techniques in the precision, recall, F1-score, and accuracy metrics for both file-based and function-based clone detection. The robust performance of CCDive emphasizes its effectiveness, reliability, accuracy, and efficiency, making it well-suited for practical applications in the real world.
{"title":"CCDive: A Deep Dive into Code Clone Detection Using Local Sequence Alignment","authors":"Yasir Glani;Luo Ping;Syed Asad Shah;Lin Ke","doi":"10.26599/TST.2024.9010075","DOIUrl":"https://doi.org/10.26599/TST.2024.9010075","url":null,"abstract":"The rapid evolution of software development has accentuated the deficiencies of prevailing code clone detection techniques. As modern applications become more complex, traditional cloning tools often struggle to detect general and large-gap clones that undergo regular modification. Such challenges pose threats to software integrity, emphasizing the critical need for improved code cloning techniques. Observing the prevailing gap, we propose an innovative code clone dive (CCDive) code cloning technique, which is designed to detect an extensive range of clones, from direct clones to the often challenging large-gap clones, thoroughly covering different categories, such as very strongly Type-III, strongly Type-III, and moderate Type-III clones. In CCDive, the fusion of a level-by-level abstraction and an innovative similarity matching algorithm ensures the recognition of clones even when nearly half the original code in the chunk has been modified. Furthermore, by integrating the Smith-Waterman local sequence alignment, the capability of CCDive to spot exact code transformation locations can be enhanced. In a comprehensive evaluation, CCDive was compared with well-known code cloning techniques. The efficacy of CCDive was measured using precision, recall, F1-score, accuracy, and efficiency. CCDive consistently surpassed other techniques in the precision, recall, F1-score, and accuracy metrics for both file-based and function-based clone detection. The robust performance of CCDive emphasizes its effectiveness, reliability, accuracy, and efficiency, making it well-suited for practical applications in the real world.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1435-1456"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908670","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553187","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2024.9010072
Xin Wang;Qingjie Zhao;Lei Wang;Wangwang Liu
Learning domain-invariant feature representations is critical to alleviate the distribution differences between training and testing domains. The existing mainstream domain generalization approaches primarily pursue to align the across-domain distributions to extract the transferable feature representations. However, these representations may be insufficient and unstable. Moreover, these networks may also undergo catastrophic forgetting because the previous learned knowledge is replaced by the new learned knowledge. To cope with these issues, we propose a novel causality-based contrastive incremental learning model for domain generalization, which mainly includes three components: (1) intra-domain causal factorization, (2) inter-domain Mahalanobis similarity metric, and (3) contrastive knowledge distillation. The model extracts intra and inter domain-invariant knowledge to improve model generalization. Specifically, we first introduce a causal factori-zation to extract intra-domain invariant knowledge. Then, we design a Mahalanobis similarity metric to extract common inter-domain invariant knowledge. Finally, we propose a contrastive knowledge distillation with exponential moving average to distill model parameters in a smooth way to preserve the previous learned knowledge and mitigate model forgetting. Extensive experiments on several domain generalization benchmarks prove that our model achieves the state-of-the-art results, which sufficiently show the effectiveness of our model.
{"title":"Causality-Based Contrastive Incremental Learning Framework for Domain Generalization","authors":"Xin Wang;Qingjie Zhao;Lei Wang;Wangwang Liu","doi":"10.26599/TST.2024.9010072","DOIUrl":"https://doi.org/10.26599/TST.2024.9010072","url":null,"abstract":"Learning domain-invariant feature representations is critical to alleviate the distribution differences between training and testing domains. The existing mainstream domain generalization approaches primarily pursue to align the across-domain distributions to extract the transferable feature representations. However, these representations may be insufficient and unstable. Moreover, these networks may also undergo catastrophic forgetting because the previous learned knowledge is replaced by the new learned knowledge. To cope with these issues, we propose a novel causality-based contrastive incremental learning model for domain generalization, which mainly includes three components: (1) intra-domain causal factorization, (2) inter-domain Mahalanobis similarity metric, and (3) contrastive knowledge distillation. The model extracts intra and inter domain-invariant knowledge to improve model generalization. Specifically, we first introduce a causal factori-zation to extract intra-domain invariant knowledge. Then, we design a Mahalanobis similarity metric to extract common inter-domain invariant knowledge. Finally, we propose a contrastive knowledge distillation with exponential moving average to distill model parameters in a smooth way to preserve the previous learned knowledge and mitigate model forgetting. Extensive experiments on several domain generalization benchmarks prove that our model achieves the state-of-the-art results, which sufficiently show the effectiveness of our model.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1636-1647"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908663","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553510","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2024.9010151
Li Huang;Bo Li;Lu Zhao
In the 5G environment, the edge computing paradigm enables service providers to deploy their service instances on distributed edge servers to serve nearby end users with extremely low latency. This boosts the emergence of modern applications, like AR/VR, online gaming, and autonomous vehicles. Existing approaches find service provision strategies under the assumption that all the user requirements are known. However, this assumption may not be true in practice and thus the effectiveness of existing approaches could be undermined. Inspired by the great success of recommender systems in various fields, we can mine users' interests in new services based on their similarities in terms of current service usage. Then, new service instances can be provisioned accordingly to better fulfil users' requirements. We formulate the problem studied in this paper as a Cost-aware Recommendation-oriented Edge Service Provision (CRESP) problem. Then, we formally model the CRESP problem as a Constrained Optimization Problem (COP). Next, we propose CRESP-O to find optimal solutions to small-scale CRESP problems. Besides, to solve large-scale CRESP problems efficiently, we propose an approximation approach named CRESP-A, which has a theoretical performance guarantee. Finally, we experimentally evaluate the performance of both CRESP-O and CRESP-A against several state-of-the-art approaches on a public testbed.
{"title":"CRESP: Cost-Aware Recommendation-Oriented Edge Service Provision","authors":"Li Huang;Bo Li;Lu Zhao","doi":"10.26599/TST.2024.9010151","DOIUrl":"https://doi.org/10.26599/TST.2024.9010151","url":null,"abstract":"In the 5G environment, the edge computing paradigm enables service providers to deploy their service instances on distributed edge servers to serve nearby end users with extremely low latency. This boosts the emergence of modern applications, like AR/VR, online gaming, and autonomous vehicles. Existing approaches find service provision strategies under the assumption that all the user requirements are known. However, this assumption may not be true in practice and thus the effectiveness of existing approaches could be undermined. Inspired by the great success of recommender systems in various fields, we can mine users' interests in new services based on their similarities in terms of current service usage. Then, new service instances can be provisioned accordingly to better fulfil users' requirements. We formulate the problem studied in this paper as a Cost-aware Recommendation-oriented Edge Service Provision (CRESP) problem. Then, we formally model the CRESP problem as a Constrained Optimization Problem (COP). Next, we propose CRESP-O to find optimal solutions to small-scale CRESP problems. Besides, to solve large-scale CRESP problems efficiently, we propose an approximation approach named CRESP-A, which has a theoretical performance guarantee. Finally, we experimentally evaluate the performance of both CRESP-O and CRESP-A against several state-of-the-art approaches on a public testbed.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1865-1884"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908596","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143535438","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}
The advancement of artificial intelligence-generated content drives the diversification of healthcare services, resulting in increased private information collection by healthcare service providers. Therefore, compliance with privacy regulations has increasingly become a paramount concern for both regulatory authorities and consumers. Privacy policies are crucial for consumers to understand how their personal information is collected, stored, and processed. In this work, we propose a privacy policy text compliance reasoning framework called FACTOR, which harnesses the power of large language models (LLMs). Since the General Data Protection Regulation (GDPR) has broad applicability, this work selects Article 13 of the GDPR as regulation requirements. FACTOR segments the privacy policy text using a sliding window strategy and employs LLM-based text entailment to assess compliance for each segment. The framework then applies a rule-based ensemble approach to aggregate the entailment results for all regulation requirements from the GDPR. Our experiments on a synthetic corpus of 388 privacy policies demonstrate the effectiveness of FACTOR. Additionally, we analyze 100 randomly selected websites offering healthcare services, revealing that nine of them lack a privacy policy altogether, while 29 have privacy policy texts that fail to meet the regulation requirements.
{"title":"A Privacy Policy Text Compliance Reasoning Framework with Large Language Models for Healthcare Services","authors":"Jintao Chen;Fan Wang;Shengye Pang;Mingshuai Chen;Meng Xi;Tiancheng Zhao;Jianwei Yin","doi":"10.26599/TST.2024.9010089","DOIUrl":"https://doi.org/10.26599/TST.2024.9010089","url":null,"abstract":"The advancement of artificial intelligence-generated content drives the diversification of healthcare services, resulting in increased private information collection by healthcare service providers. Therefore, compliance with privacy regulations has increasingly become a paramount concern for both regulatory authorities and consumers. Privacy policies are crucial for consumers to understand how their personal information is collected, stored, and processed. In this work, we propose a privacy policy text compliance reasoning framework called FACTOR, which harnesses the power of large language models (LLMs). Since the General Data Protection Regulation (GDPR) has broad applicability, this work selects Article 13 of the GDPR as regulation requirements. FACTOR segments the privacy policy text using a sliding window strategy and employs LLM-based text entailment to assess compliance for each segment. The framework then applies a rule-based ensemble approach to aggregate the entailment results for all regulation requirements from the GDPR. Our experiments on a synthetic corpus of 388 privacy policies demonstrate the effectiveness of FACTOR. Additionally, we analyze 100 randomly selected websites offering healthcare services, revealing that nine of them lack a privacy policy altogether, while 29 have privacy policy texts that fail to meet the regulation requirements.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1831-1845"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908666","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143535484","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2024.9010214
Guangfeng Li;Jian Sun;Donglei Du;Xiaoyan Zhang
We consider the Max Directed 3-Section problem, which is closely connected to other well-known graph partition problems, such as Max Cut and Max Bisection. Given an arc-weighted directed graph, the goal of the Max Directed 3-Section problem is to partition the vertex set into three disjoint subsets with equal size, while maximizing the total weight of arcs crossing different vertex subsets. By combining the Lasserre hierarchy with the random hyperplane rounding strategy, we propose a polynomial-time algorithm with approximation ratio of 0.489.
{"title":"A Lasserre SDP Rounding Approximation Algorithm for Max Directed 3-Section","authors":"Guangfeng Li;Jian Sun;Donglei Du;Xiaoyan Zhang","doi":"10.26599/TST.2024.9010214","DOIUrl":"https://doi.org/10.26599/TST.2024.9010214","url":null,"abstract":"We consider the Max Directed 3-Section problem, which is closely connected to other well-known graph partition problems, such as Max Cut and Max Bisection. Given an arc-weighted directed graph, the goal of the Max Directed 3-Section problem is to partition the vertex set into three disjoint subsets with equal size, while maximizing the total weight of arcs crossing different vertex subsets. By combining the Lasserre hierarchy with the random hyperplane rounding strategy, we propose a polynomial-time algorithm with approximation ratio of 0.489.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1885-1896"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908669","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143535532","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2024.9010062
Wenfeng Yi;Wenhan Wu;Maoyin Chen;Xiaoping Zheng
Autonomous machines (AMs) are poised to possess human-like moral cognition, yet their morality is often pre-programmed for safety. This raises the question of whether the morality intended by programmers aligns with their actions during actual operation, a crucial consideration for a future society with both humans and AMs. Investigating this, we use a micro-robot swarm in a simulated fire scenario, with 180 participants, including 102 robot programmers, completing moral questionnaires and participating in virtual escape trials. These exercises mirror common societal moral dilemmas. Our comparative analysis reveals a “morality gap” between programming presets and real-time operation, primarily influenced by uncertainty about the future and heightened by external pressures, especially social punishment. This discrepancy suggests that operational morality can diverge from programmed intentions, underlining the need for careful AM design to foster a collaborative and efficient society.
{"title":"Human Morality Difference when Programming and Actually Operating Autonomous Machines","authors":"Wenfeng Yi;Wenhan Wu;Maoyin Chen;Xiaoping Zheng","doi":"10.26599/TST.2024.9010062","DOIUrl":"https://doi.org/10.26599/TST.2024.9010062","url":null,"abstract":"Autonomous machines (AMs) are poised to possess human-like moral cognition, yet their morality is often pre-programmed for safety. This raises the question of whether the morality intended by programmers aligns with their actions during actual operation, a crucial consideration for a future society with both humans and AMs. Investigating this, we use a micro-robot swarm in a simulated fire scenario, with 180 participants, including 102 robot programmers, completing moral questionnaires and participating in virtual escape trials. These exercises mirror common societal moral dilemmas. Our comparative analysis reveals a “morality gap” between programming presets and real-time operation, primarily influenced by uncertainty about the future and heightened by external pressures, especially social punishment. This discrepancy suggests that operational morality can diverge from programmed intentions, underlining the need for careful AM design to foster a collaborative and efficient society.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1648-1658"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908660","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553222","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2024.9010083
Edris Khezri;Hiwa Hassanzadeh;Rebaz Othman Yahya;Mahdi Mir
Due to their diverse applications, including safety, welfare, and improving traffic efficiency, inter-vehicle ad-hoc networks have been extensively studied. Globally, road congestion, accidents, fuel consumption, and environmental pollution caused by the large number of vehicles have become serious problems that have caused a lot of human and financial losses. Intelligent transportation systems (ITS) have introduced VANETs in order to overcome these problems. In vehicular ad hoc networks (VANETs), vehicles equipped with wireless interfaces can communicate with other vehicles and with fixed roadside equipment via mobile ad hoc networks (MANETs). Messages are transmitted over open wireless channels in VANETs. Malicious nodes target these networks to protect them from various attacks, such as interference, eavesdropping, spoofing, denial of service, Sybil, black holes, worm holes, gray holes, etc. Security of VANETs is therefore one of the most significant issues. Security issues, attacks, attackers, and secure routing protocols in VANETs are discussed in this article, as well as available solutions to solve security issues.
{"title":"Security Challenges in Internet of Vehicles (IoV) for ITS: A Survey","authors":"Edris Khezri;Hiwa Hassanzadeh;Rebaz Othman Yahya;Mahdi Mir","doi":"10.26599/TST.2024.9010083","DOIUrl":"https://doi.org/10.26599/TST.2024.9010083","url":null,"abstract":"Due to their diverse applications, including safety, welfare, and improving traffic efficiency, inter-vehicle ad-hoc networks have been extensively studied. Globally, road congestion, accidents, fuel consumption, and environmental pollution caused by the large number of vehicles have become serious problems that have caused a lot of human and financial losses. Intelligent transportation systems (ITS) have introduced VANETs in order to overcome these problems. In vehicular ad hoc networks (VANETs), vehicles equipped with wireless interfaces can communicate with other vehicles and with fixed roadside equipment via mobile ad hoc networks (MANETs). Messages are transmitted over open wireless channels in VANETs. Malicious nodes target these networks to protect them from various attacks, such as interference, eavesdropping, spoofing, denial of service, Sybil, black holes, worm holes, gray holes, etc. Security of VANETs is therefore one of the most significant issues. Security issues, attacks, attackers, and secure routing protocols in VANETs are discussed in this article, as well as available solutions to solve security issues.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1700-1723"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908658","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553509","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2023.9010096
Dejuan Li;James A. Esquivel
This paper introduces a novel trust-aware hybrid recommendation framework that combines Locality-Sensitive Hashing (LSH) with the trust information in social networks, aiming to provide efficient and effective recommendations. Unlike traditional recommender systems which often overlook the critical influence of user trust, our proposed approach infuses trust metrics to better approximate user preferences. The LSH, with its intrinsic advantage in handling high-dimensional data and computational efficiency, is applied to expedite the process of finding similar items or users. We innovatively adapt LSH to form trust-aware buckets, encapsulating both trust and similarity information. These enhancements mitigate the sparsity and scalability issues usually found in existing recommender systems. Experimental results on a real-world dataset confirm the superiority of our approach in terms of recommendation quality and computational performance. The paper further discusses potential applications and future directions of the trust-aware hybrid recommendation with LSH.
{"title":"Trust-Aware Hybrid Collaborative Recommendation with Locality-Sensitive Hashing","authors":"Dejuan Li;James A. Esquivel","doi":"10.26599/TST.2023.9010096","DOIUrl":"https://doi.org/10.26599/TST.2023.9010096","url":null,"abstract":"This paper introduces a novel trust-aware hybrid recommendation framework that combines Locality-Sensitive Hashing (LSH) with the trust information in social networks, aiming to provide efficient and effective recommendations. Unlike traditional recommender systems which often overlook the critical influence of user trust, our proposed approach infuses trust metrics to better approximate user preferences. The LSH, with its intrinsic advantage in handling high-dimensional data and computational efficiency, is applied to expedite the process of finding similar items or users. We innovatively adapt LSH to form trust-aware buckets, encapsulating both trust and similarity information. These enhancements mitigate the sparsity and scalability issues usually found in existing recommender systems. Experimental results on a real-world dataset confirm the superiority of our approach in terms of recommendation quality and computational performance. The paper further discusses potential applications and future directions of the trust-aware hybrid recommendation with LSH.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1421-1434"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908673","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553183","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2024.9010058
Junsan Zhang;Ming Cheng;Xiangyang Li;Xiuxuan Shen;Yuxue Liu;Yao Wan
In medical X-ray images, multiple abnormalities may occur frequently. However, existing report generation methods cannot efficiently extract all abnormal features, resulting in incomplete disease diagnoses when generating diagnostic reports. In real medical scenarios, there are co-occurrence relations among multiple diseases. If such co-occurrence relations are mined and integrated into the feature extraction process, the issue of missing abnormal features may be addressed. Inspired by this observation, we propose a novel method to improve the extraction of abnormal features in images through joint probability graph reasoning. Specifically, to reveal the co-occurrence relations among multiple diseases, we conduct statistical analyses on the dataset, and extract disease relationships into a probability map. Subsequently, we devise a graph reasoning network for conducting correlation-based reasoning over the features of medical images, which can facilitate the acquisition of more abnormal features. Furthermore, we introduce a gating mechanism focused on cross-modal features fusion into the current text generation model. This optimization substantially improves the model's capabilities to learn and fuse information from two distinct modalities-medical images and texts. Experimental results on the IU-X-Ray and MIMIC-CXR datasets demonstrate that our approach outperforms previous state-of-the-art methods, exhibiting the ability to generate higher quality medical image reports.
{"title":"Generating Medical Report via Joint Probability Graph Reasoning","authors":"Junsan Zhang;Ming Cheng;Xiangyang Li;Xiuxuan Shen;Yuxue Liu;Yao Wan","doi":"10.26599/TST.2024.9010058","DOIUrl":"https://doi.org/10.26599/TST.2024.9010058","url":null,"abstract":"In medical X-ray images, multiple abnormalities may occur frequently. However, existing report generation methods cannot efficiently extract all abnormal features, resulting in incomplete disease diagnoses when generating diagnostic reports. In real medical scenarios, there are co-occurrence relations among multiple diseases. If such co-occurrence relations are mined and integrated into the feature extraction process, the issue of missing abnormal features may be addressed. Inspired by this observation, we propose a novel method to improve the extraction of abnormal features in images through joint probability graph reasoning. Specifically, to reveal the co-occurrence relations among multiple diseases, we conduct statistical analyses on the dataset, and extract disease relationships into a probability map. Subsequently, we devise a graph reasoning network for conducting correlation-based reasoning over the features of medical images, which can facilitate the acquisition of more abnormal features. Furthermore, we introduce a gating mechanism focused on cross-modal features fusion into the current text generation model. This optimization substantially improves the model's capabilities to learn and fuse information from two distinct modalities-medical images and texts. Experimental results on the IU-X-Ray and MIMIC-CXR datasets demonstrate that our approach outperforms previous state-of-the-art methods, exhibiting the ability to generate higher quality medical image reports.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1685-1699"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908661","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553184","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}
Pub Date : 2025-03-03DOI: 10.26599/TST.2023.9010130
Tengfei Fan;Wenmin Lin
A joint green-edge computing idea is now realized in practice with the help of intelligent infrastructure for modern sport venues, based on Internet of Things (IoT) platforms and Cyber-Physical Systems (CPS). To monitor their sports actions, athletes need smart environments. Using edge-enabled low-cost and low-power sensors, such as infrared monitoring systems that analyze thermal information, this environment should alert to possible physical damages. Early recognition of sports injuries and joint injuries can usually prevent athletes from pain and missing exercise. One of the most efficient methods for identifying pain and movement problems is to monitor the energy emitted by lower limb injuries. By analyzing thermal images of the lower body parts, this research attempts to automatically identify sports injuries. The thermal image is first isolated from the region of interest. Convolutional structures are applied to identify lesions using a newly developed and optimized method. The performance of the classifier is performed with the possibility of deep learning by pruning the features, to reduce the computational complexity and improve the accuracy, and a model has been developed based on the classification of sports injuries in binary mode (i.e., whether the lesions are present or not) and multiclass mode (i.e., the severity of sports injuries) resulted in optimal results. Thermal images show the different states of joints, including lesions caused by various sports in the lower limbs. This model could provide the ability of solving uncertainty of answers, repeatability, and convergence towards minimum error. As compared to conventional feature extraction and classification approaches, the outputs are more acceptable. By taking advantage of the K-fold cross-validation method, the average error of the proposed method to detect the severity of damage is less than 2.22%.
{"title":"Deep Learning-Based Thermal Imaging Analysis to Diagnose Abnormalities in Sports Buildings: Smart Cyber-Physical Monitoring Sensors at the Edge","authors":"Tengfei Fan;Wenmin Lin","doi":"10.26599/TST.2023.9010130","DOIUrl":"https://doi.org/10.26599/TST.2023.9010130","url":null,"abstract":"A joint green-edge computing idea is now realized in practice with the help of intelligent infrastructure for modern sport venues, based on Internet of Things (IoT) platforms and Cyber-Physical Systems (CPS). To monitor their sports actions, athletes need smart environments. Using edge-enabled low-cost and low-power sensors, such as infrared monitoring systems that analyze thermal information, this environment should alert to possible physical damages. Early recognition of sports injuries and joint injuries can usually prevent athletes from pain and missing exercise. One of the most efficient methods for identifying pain and movement problems is to monitor the energy emitted by lower limb injuries. By analyzing thermal images of the lower body parts, this research attempts to automatically identify sports injuries. The thermal image is first isolated from the region of interest. Convolutional structures are applied to identify lesions using a newly developed and optimized method. The performance of the classifier is performed with the possibility of deep learning by pruning the features, to reduce the computational complexity and improve the accuracy, and a model has been developed based on the classification of sports injuries in binary mode (i.e., whether the lesions are present or not) and multiclass mode (i.e., the severity of sports injuries) resulted in optimal results. Thermal images show the different states of joints, including lesions caused by various sports in the lower limbs. This model could provide the ability of solving uncertainty of answers, repeatability, and convergence towards minimum error. As compared to conventional feature extraction and classification approaches, the outputs are more acceptable. By taking advantage of the K-fold cross-validation method, the average error of the proposed method to detect the severity of damage is less than 2.22%.","PeriodicalId":48690,"journal":{"name":"Tsinghua Science and Technology","volume":"30 4","pages":"1457-1473"},"PeriodicalIF":6.6,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10908671","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553180","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：