Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00267
P. Kasnesis, Lazaros Toumanidis, A. Burrello, Christos Chatzigeorgiou, C. Patrikakis
Nowadays, Hearth Rate (HR) monitoring is a key feature of almost all wrist-worn devices exploiting photoplethysmography (PPG) sensors. However, arm movements affect the performance of PPG-based HR tracking. This issue is usually addressed by fusing the PPG signal with data produced by inertial measurement units. Thus, deep learning algorithms have been proposed, but they are considered too complex to deploy on wearable devices and lack the explainability of results. In this work, we present a new deep learning model, PULSE, which exploits temporal convolutions and feature-level multi-head cross-attention to improve sensor fusion’s effectiveness and achieve a step towards explainability. We evaluate the performance of PULSE on three publicly available datasets, reducing the mean absolute error by 7.56% on the most extensive available dataset, PPG-DaLiA. Finally, we demonstrate the explainability of PULSE and the benefits of applying attention modules to PPG and motion data.
{"title":"Feature-Level Cross-Attentional PPG and Motion Signal Fusion for Heart Rate Estimation","authors":"P. Kasnesis, Lazaros Toumanidis, A. Burrello, Christos Chatzigeorgiou, C. Patrikakis","doi":"10.1109/COMPSAC57700.2023.00267","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00267","url":null,"abstract":"Nowadays, Hearth Rate (HR) monitoring is a key feature of almost all wrist-worn devices exploiting photoplethysmography (PPG) sensors. However, arm movements affect the performance of PPG-based HR tracking. This issue is usually addressed by fusing the PPG signal with data produced by inertial measurement units. Thus, deep learning algorithms have been proposed, but they are considered too complex to deploy on wearable devices and lack the explainability of results. In this work, we present a new deep learning model, PULSE, which exploits temporal convolutions and feature-level multi-head cross-attention to improve sensor fusion’s effectiveness and achieve a step towards explainability. We evaluate the performance of PULSE on three publicly available datasets, reducing the mean absolute error by 7.56% on the most extensive available dataset, PPG-DaLiA. Finally, we demonstrate the explainability of PULSE and the benefits of applying attention modules to PPG and motion data.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"62 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127005238","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00270
Itsuki Matsunaga, Yuto Kosugi, Hangli Ge, Takashi Michikata, N. Koshizuka
Long-term traffic prediction is essential for both road managers and users to prepare for future congestion. However, most existing studies have only focused on short-term prediction. Moreover, few studies have effectively incorporated external data into long-term traffic prediction, even though traffic conditions are complexly influenced by various spatiotemporal factors. In this paper, we propose a novel method that utilizes online search log data for long-term traffic prediction on expressways. Online search logs reflect drivers’ travel intentions and external factors, such as weather conditions and events, which cannot be represented by historical traffic data. Based on a new analysis of the correlation between online search log data and real-world traffic, we use online search log data as potential future traffic volume in an LSTM-based encoder-decoder model. Experiments using a real-world dataset on an expressway known for frequent congestion show that the use of online search log data improves the metrics of MAE, RMSE, and MAPE in next-day traffic volume prediction by 8.1%, 12.5%, and 7.2% on average, respectively. Similarly, in speed prediction, the MAE, RMSE, and MAPE are reduced by 3.7%, 2.1%, and 11.8%, respectively. It is also shown that online search log data is particularly effective in predicting irregular congestion caused by sudden increases in traffic demand.
{"title":"Improving Long-Term Traffic Prediction with Online Search Log Data","authors":"Itsuki Matsunaga, Yuto Kosugi, Hangli Ge, Takashi Michikata, N. Koshizuka","doi":"10.1109/COMPSAC57700.2023.00270","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00270","url":null,"abstract":"Long-term traffic prediction is essential for both road managers and users to prepare for future congestion. However, most existing studies have only focused on short-term prediction. Moreover, few studies have effectively incorporated external data into long-term traffic prediction, even though traffic conditions are complexly influenced by various spatiotemporal factors. In this paper, we propose a novel method that utilizes online search log data for long-term traffic prediction on expressways. Online search logs reflect drivers’ travel intentions and external factors, such as weather conditions and events, which cannot be represented by historical traffic data. Based on a new analysis of the correlation between online search log data and real-world traffic, we use online search log data as potential future traffic volume in an LSTM-based encoder-decoder model. Experiments using a real-world dataset on an expressway known for frequent congestion show that the use of online search log data improves the metrics of MAE, RMSE, and MAPE in next-day traffic volume prediction by 8.1%, 12.5%, and 7.2% on average, respectively. Similarly, in speed prediction, the MAE, RMSE, and MAPE are reduced by 3.7%, 2.1%, and 11.8%, respectively. It is also shown that online search log data is particularly effective in predicting irregular congestion caused by sudden increases in traffic demand.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"93 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130564744","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00022
Theodora Adufu, Yoonhee Kim
Cache management is a significant aspect of executing applications on GPUs. With the advancements in GPU architecture, issues such as data reuse, cache line eviction and data residency are to be considered for optimal performance. Frequency of data access from global memory has significant impacts on the performance of the application with increased latencies. However, the L2 cache data residency feature by NVIDIA promises to reduce the overheads associated with frequent data accesses. Through the information extracted from static profiling analysis, we quantitatively analyzed the frequency of data reuse by threads to determine whether an application has frequent data accesses or not. We also estimated the size of access policy window from which persistent data should be cached to avoid stalling of warps. Also with our proposed approach, we observed that L1 cache load throughput increased by 2.75% for GEMM, 0.33% for 2DConv St and 0.46% for 2DConv Large respectively as data was resident in the L2 cache.
{"title":"L2 Cache Access Pattern Analysis using Static Profiling of an Application","authors":"Theodora Adufu, Yoonhee Kim","doi":"10.1109/COMPSAC57700.2023.00022","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00022","url":null,"abstract":"Cache management is a significant aspect of executing applications on GPUs. With the advancements in GPU architecture, issues such as data reuse, cache line eviction and data residency are to be considered for optimal performance. Frequency of data access from global memory has significant impacts on the performance of the application with increased latencies. However, the L2 cache data residency feature by NVIDIA promises to reduce the overheads associated with frequent data accesses. Through the information extracted from static profiling analysis, we quantitatively analyzed the frequency of data reuse by threads to determine whether an application has frequent data accesses or not. We also estimated the size of access policy window from which persistent data should be cached to avoid stalling of warps. Also with our proposed approach, we observed that L1 cache load throughput increased by 2.75% for GEMM, 0.33% for 2DConv St and 0.46% for 2DConv Large respectively as data was resident in the L2 cache.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"179 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123839100","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00244
Hüseyin Yapici, Hasan Sözer
This paper proposes a new metric, namely the coevolution index (CEI), for measuring the relative evolutionary coupling of modules of a software system. CEI is inspired by the h-index, which is a popular metric used for measuring the productivity and citation impact of scholars and scientists. CEI of a module is equal to n, which is the number of times it is modified together with at least n other modules of the system. We develop a script that can calculate CEI for source files in a code repository. We analyze the repository of 4 software systems. Source files that are subject to a high number of changes to address issues tend to have high CEI scores. CEI also reflects a relative footprint in maintenance efforts by definition. Hence, it can help in tracking technical debt interest and focusing the refactoring efforts for improving maintainability and reusability.
{"title":"Coevolution Index: A Metric for Tracking Evolutionary Coupling","authors":"Hüseyin Yapici, Hasan Sözer","doi":"10.1109/COMPSAC57700.2023.00244","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00244","url":null,"abstract":"This paper proposes a new metric, namely the coevolution index (CEI), for measuring the relative evolutionary coupling of modules of a software system. CEI is inspired by the h-index, which is a popular metric used for measuring the productivity and citation impact of scholars and scientists. CEI of a module is equal to n, which is the number of times it is modified together with at least n other modules of the system. We develop a script that can calculate CEI for source files in a code repository. We analyze the repository of 4 software systems. Source files that are subject to a high number of changes to address issues tend to have high CEI scores. CEI also reflects a relative footprint in maintenance efforts by definition. Hence, it can help in tracking technical debt interest and focusing the refactoring efforts for improving maintainability and reusability.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124030652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00029
Salim Saay, A. Norta
Collaboration between schools at a cross-national level requires a secure and trustable system that respects the rules, and policies of the parties, specifically the General Data Protection Regulation(GDPR) needs to be placed in the system. The GDPR, or similar laws and the curricula differ in respective. Students who want to take any course at another school abroad cannot easily sign in and access the study content, even if that school wishes to share it. Still, this is what society, industry, and education ministries in different countries wish. The schoolgirls in Afghanistan who wish to study online in the girl’s school of Ireland are specifically considered as a case study in this paper. We use a case study research method in which the existing online learning tools are analysed that are used in Irshad High School in Afghanistan and Coláiste Nano Nagle School in Limerick, Ireland. We furthermore prototype the architecture of a system that provides a secure collaboration platform. Thus, this collaboration platform we develop can be adopted at various levels of education for any country with the goal to achieve a long-term effect on the flexibility of the respective education systems to yield a globalization of education. We use the internet, mobile and educational platforms, and a secure broker (interface) that provides the tools for the schools’ collaboration, including the sharing of resources, an exchange of experience and enabling access to education worldwide.
{"title":"Requirements for an international educational collaboration system architecture, a case study: Coláiste Nano Nagle School in Limerick, Ireland, and Irshad High School in Kabul, Afghanistan","authors":"Salim Saay, A. Norta","doi":"10.1109/COMPSAC57700.2023.00029","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00029","url":null,"abstract":"Collaboration between schools at a cross-national level requires a secure and trustable system that respects the rules, and policies of the parties, specifically the General Data Protection Regulation(GDPR) needs to be placed in the system. The GDPR, or similar laws and the curricula differ in respective. Students who want to take any course at another school abroad cannot easily sign in and access the study content, even if that school wishes to share it. Still, this is what society, industry, and education ministries in different countries wish. The schoolgirls in Afghanistan who wish to study online in the girl’s school of Ireland are specifically considered as a case study in this paper. We use a case study research method in which the existing online learning tools are analysed that are used in Irshad High School in Afghanistan and Coláiste Nano Nagle School in Limerick, Ireland. We furthermore prototype the architecture of a system that provides a secure collaboration platform. Thus, this collaboration platform we develop can be adopted at various levels of education for any country with the goal to achieve a long-term effect on the flexibility of the respective education systems to yield a globalization of education. We use the internet, mobile and educational platforms, and a secure broker (interface) that provides the tools for the schools’ collaboration, including the sharing of resources, an exchange of experience and enabling access to education worldwide.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114684262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00034
Rong Liu, Zemao Chen, Jiayi Liu
In recent years, cyber threats have significantly increased in sophistication and targeted nature. Traditional security measures often prove inadequate in detecting malicious activities. Intrusion Detection Systems (IDS) can mitigate these threats by monitoring and alerting administrators to suspicious activities. However, the large volume and high dimensionality of network traffic data can pose a challenge for IDS, as irrelevant and redundant features can reduce the effectiveness of detection. Additionally, many machine learning-based IDS rely on individual base classifiers, which can lack robustness and may not perform well in varying situations. To address these issues, this paper proposes a hybrid IDS that combines feature selection and voting classifier techniques. The proposed model utilizes an improved binary Pigeon-Inspired Optimization algorithm and the Minimal-Redundancy-Maximal-Relevance algorithm for feature selection, and a voting classifier incorporating Random Forest, K-Nearest Neighbors, and XGBoost to classify network traffic. The model has been evaluated on three popular datasets: KDDCUP99, NLS-KDD and CIC-IDS2017. The proposed method demonstrates superior performance in terms of Accuracy, Precision, Recall, F1-score, and False Positive Rate when compared to several machine learning and deep learning models.
{"title":"A Hybrid Intrusion Detection System Based on Feature Selection and Voting Classifier","authors":"Rong Liu, Zemao Chen, Jiayi Liu","doi":"10.1109/COMPSAC57700.2023.00034","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00034","url":null,"abstract":"In recent years, cyber threats have significantly increased in sophistication and targeted nature. Traditional security measures often prove inadequate in detecting malicious activities. Intrusion Detection Systems (IDS) can mitigate these threats by monitoring and alerting administrators to suspicious activities. However, the large volume and high dimensionality of network traffic data can pose a challenge for IDS, as irrelevant and redundant features can reduce the effectiveness of detection. Additionally, many machine learning-based IDS rely on individual base classifiers, which can lack robustness and may not perform well in varying situations. To address these issues, this paper proposes a hybrid IDS that combines feature selection and voting classifier techniques. The proposed model utilizes an improved binary Pigeon-Inspired Optimization algorithm and the Minimal-Redundancy-Maximal-Relevance algorithm for feature selection, and a voting classifier incorporating Random Forest, K-Nearest Neighbors, and XGBoost to classify network traffic. The model has been evaluated on three popular datasets: KDDCUP99, NLS-KDD and CIC-IDS2017. The proposed method demonstrates superior performance in terms of Accuracy, Precision, Recall, F1-score, and False Positive Rate when compared to several machine learning and deep learning models.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"110 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126356690","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00031
Bin Liu, Xiao-Xiong Bai, Xinyue Su, Chenxi Song, Zhuohan Yao, Xing Wei, Haixi Zhang
Early detection of apple pests and diseases is beneficial to ensure the healthy development of the apple industry. However, previous detectors typically suffer from low accuracy and poor timeliness, limiting their further application in real scenarios as well as complex backgrounds. A real-time early apple leaf pests and diseases detection model is proposed in this paper, dubbed DAC-PPYOLOE+. Firstly, an efficient adaptive feature fusion strategy is utilized to improve the detection capability of the model under complex backgrounds. Meanwhile, a new ESPBlock with dilated convolution implemented by depthwise separable convolution is designed, which greatly reduces the number of parameters and enhances the adaptability to different scale targets. Furthermore, a novel skip information transmission structure is proposed to fully exploit the information of deep and shallow feature maps, which is specifically used for small target detection. Compared to the baseline model, DAC-PPYOLOE + has a smaller number of parameters and achieves 44.9% AP on the COCO test, which is 3.6% higher, and the inference speed is 10.0 ms, faster about 2 ms. Experimental results of comparison with various advanced detection methods show that the proposed model outperforms various advanced algorithms in handling early apple leaf pests and diseases detection tasks, indicating that DAC-PPYOLOE+ provides effective technical support for real-time and accurate detection of early apple leaf pests and diseases under complex backgrounds.
{"title":"DAC-PPYOLOE+: A Lightweight Real-time Detection Model for Early Apple Leaf Pests and Diseases under Complex Background","authors":"Bin Liu, Xiao-Xiong Bai, Xinyue Su, Chenxi Song, Zhuohan Yao, Xing Wei, Haixi Zhang","doi":"10.1109/COMPSAC57700.2023.00031","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00031","url":null,"abstract":"Early detection of apple pests and diseases is beneficial to ensure the healthy development of the apple industry. However, previous detectors typically suffer from low accuracy and poor timeliness, limiting their further application in real scenarios as well as complex backgrounds. A real-time early apple leaf pests and diseases detection model is proposed in this paper, dubbed DAC-PPYOLOE+. Firstly, an efficient adaptive feature fusion strategy is utilized to improve the detection capability of the model under complex backgrounds. Meanwhile, a new ESPBlock with dilated convolution implemented by depthwise separable convolution is designed, which greatly reduces the number of parameters and enhances the adaptability to different scale targets. Furthermore, a novel skip information transmission structure is proposed to fully exploit the information of deep and shallow feature maps, which is specifically used for small target detection. Compared to the baseline model, DAC-PPYOLOE + has a smaller number of parameters and achieves 44.9% AP on the COCO test, which is 3.6% higher, and the inference speed is 10.0 ms, faster about 2 ms. Experimental results of comparison with various advanced detection methods show that the proposed model outperforms various advanced algorithms in handling early apple leaf pests and diseases detection tasks, indicating that DAC-PPYOLOE+ provides effective technical support for real-time and accurate detection of early apple leaf pests and diseases under complex backgrounds.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"79 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128109573","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00238
Paramita Basak Upama, Anushka Kolli, Hansika Kolli, Subarna Alam, Mohammad Syam, H. Shahriar, S. Ahamed
Quantum machine learning (QML) in the field of disease detection and prediction use quantum computing techniques and algorithms to analyze and classify large datasets of medical information, by identifying subtle patterns and predict the occurrence or progression of diseases. It involves applying machine learning techniques to data from biological and medical research, such as-genomic and proteomic data, medical imaging, electronic health records, and clinical trial data, using quantum computing algorithms and architectures to perform these analyses more efficiently and accurately than classical computing methods. This approach has the potential to provide new insights into complex biological systems and facilitate the development of more effective treatments and personalized medicine. In this paper, a systematic review of the use of QML algorithms has been conducted, which focuses on the detection and prediction of diseases among patients. The current essence of the field along with the challenges and limitations of current works have also been discussed. After evaluating the implemented and proposed methods of data analysis, algorithm development, usefulness and efficiency of the system in various disease detection and prediction, a recommendation was made on the open research scopes in this field at the end of the paper.
{"title":"Quantum Machine Learning in Disease Detection and Prediction: a survey of applications and future possibilities","authors":"Paramita Basak Upama, Anushka Kolli, Hansika Kolli, Subarna Alam, Mohammad Syam, H. Shahriar, S. Ahamed","doi":"10.1109/COMPSAC57700.2023.00238","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00238","url":null,"abstract":"Quantum machine learning (QML) in the field of disease detection and prediction use quantum computing techniques and algorithms to analyze and classify large datasets of medical information, by identifying subtle patterns and predict the occurrence or progression of diseases. It involves applying machine learning techniques to data from biological and medical research, such as-genomic and proteomic data, medical imaging, electronic health records, and clinical trial data, using quantum computing algorithms and architectures to perform these analyses more efficiently and accurately than classical computing methods. This approach has the potential to provide new insights into complex biological systems and facilitate the development of more effective treatments and personalized medicine. In this paper, a systematic review of the use of QML algorithms has been conducted, which focuses on the detection and prediction of diseases among patients. The current essence of the field along with the challenges and limitations of current works have also been discussed. After evaluating the implemented and proposed methods of data analysis, algorithm development, usefulness and efficiency of the system in various disease detection and prediction, a recommendation was made on the open research scopes in this field at the end of the paper.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128169209","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00180
Fernando Martinez, Yijun Zhao
Inspired by the success of various visual attention techniques in computer vision, we introduce a novel method for integrating multiple attention mechanisms to boost model performance. Our approach involves augmenting a base model with a Parallel Visual Attention Encoder (PVAE) branch, which concurrently employs two different attention modules (modified large kernel attention and modified convolutional block attention) to capture essential visual features. To reduce the training cost incurred by these additional components, we apply an encoder for efficient feature extraction and dimensionality reduction before applying the attention modules. The proposed PVAE architecture can be combined with cutting-edge models (e.g., EfficientNet, ResNet, DenseNet, etc.) to create a Parallel Visual Attention Network (PVAN). We evaluate the efficacy of our approach by devising a PVAN with EfficientNet as the base model for the task of classifying dog breeds. Our experimental results demonstrate the effectiveness of the proposed hybrid visual attention architecture, which achieves superior performance compared to the base model and models with a single attention mechanism. We further present an interactive web application developed for the general public to identify dog breeds using their photographs to test our model’s performance in real-life scenarios.
{"title":"Integrating Multiple Visual Attention Mechanisms in Deep Neural Networks","authors":"Fernando Martinez, Yijun Zhao","doi":"10.1109/COMPSAC57700.2023.00180","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00180","url":null,"abstract":"Inspired by the success of various visual attention techniques in computer vision, we introduce a novel method for integrating multiple attention mechanisms to boost model performance. Our approach involves augmenting a base model with a Parallel Visual Attention Encoder (PVAE) branch, which concurrently employs two different attention modules (modified large kernel attention and modified convolutional block attention) to capture essential visual features. To reduce the training cost incurred by these additional components, we apply an encoder for efficient feature extraction and dimensionality reduction before applying the attention modules. The proposed PVAE architecture can be combined with cutting-edge models (e.g., EfficientNet, ResNet, DenseNet, etc.) to create a Parallel Visual Attention Network (PVAN). We evaluate the efficacy of our approach by devising a PVAN with EfficientNet as the base model for the task of classifying dog breeds. Our experimental results demonstrate the effectiveness of the proposed hybrid visual attention architecture, which achieves superior performance compared to the base model and models with a single attention mechanism. We further present an interactive web application developed for the general public to identify dog breeds using their photographs to test our model’s performance in real-life scenarios.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"135 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125682260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-06-01DOI: 10.1109/COMPSAC57700.2023.00261
F. Siewe
The Unified Modeling Language (UML) is the industrial de-facto standard for designing systems. It has been used widely in many industrial applications. However, the lack of formal semantics for UML makes it unsuitable for formal verification. As such, UML is limited when it comes to the design of safety/security critical systems where faults can cause damages to people, properties, or the environment. This paper proposes an attempt to define a formal semantics for the UML activity diagrams. An algorithm is proposed that translates an activity diagram into a process in a Calculus of Context-aware Ambients (CCA). This process can then be formally analysed using the tool support for CCA. Hence, errors can be detected and fixed early during the system development life-cycle. The pragmatics of the proposed approach is demonstrated using a case study in e-commerce.
{"title":"Towards the Formal Analysis of UML Activity Diagrams in a Calculus of Context-aware Ambients","authors":"F. Siewe","doi":"10.1109/COMPSAC57700.2023.00261","DOIUrl":"https://doi.org/10.1109/COMPSAC57700.2023.00261","url":null,"abstract":"The Unified Modeling Language (UML) is the industrial de-facto standard for designing systems. It has been used widely in many industrial applications. However, the lack of formal semantics for UML makes it unsuitable for formal verification. As such, UML is limited when it comes to the design of safety/security critical systems where faults can cause damages to people, properties, or the environment. This paper proposes an attempt to define a formal semantics for the UML activity diagrams. An algorithm is proposed that translates an activity diagram into a process in a Calculus of Context-aware Ambients (CCA). This process can then be formally analysed using the tool support for CCA. Hence, errors can be detected and fixed early during the system development life-cycle. The pragmatics of the proposed approach is demonstrated using a case study in e-commerce.","PeriodicalId":296288,"journal":{"name":"2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130029840","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","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：