Pub Date : 2023-06-01DOI: 10.1109/IS3C57901.2023.00067
Liang-Wei Ouyang, J. Mayeda, C. Sweeney, G. Somasundaram, D. Lie, Jerry Lopez
This paper presents a design of a broadband millimeter-wave (mm-Wave) low noise amplifier (LNA) designed in a 22 nm fully-depleted silicon-on-insulator (FD-SOI) CMOS technology. The post-layout parasitic extracted (PEX) simulations suggest the LNA has a 3-dB bandwidth (BW) from 16.9 – 41.8 GHz and a fractional bandwidth (FBW) of 84.8 %, covering the key frequency bands within the mm-Wave 5G FR2 band, with its noise figure (NF) ranging from 2.9 – 4.1 dB, and its input-referred 1-dB compression point (IP1dB) of −19.4 dBm at 28 GHz with 15.8 mW DC power consumption. Using the FOM (figure-of-merit) developed from Ref. [1] for broadband LNAs (FOM $equivmathbf{2 0}timeslog((Gain[V / V]times BW[GHz])/(P_{DC}[mW]times(F-1))$, this LNA achieves a competitive FOM among reported mm-Wave LNAs in literature [1–7].
提出了一种基于22 nm全耗尽绝缘体上硅(FD-SOI) CMOS技术的宽带毫米波(mm-Wave)低噪声放大器(LNA)的设计。布局后寄生提取(PEX)仿真表明,LNA在16.9 ~ 41.8 GHz范围内具有3db带宽(BW),分数带宽(FBW)为84.8 GHz %, covering the key frequency bands within the mm-Wave 5G FR2 band, with its noise figure (NF) ranging from 2.9 – 4.1 dB, and its input-referred 1-dB compression point (IP1dB) of −19.4 dBm at 28 GHz with 15.8 mW DC power consumption. Using the FOM (figure-of-merit) developed from Ref. [1] for broadband LNAs (FOM $equivmathbf{2 0}timeslog((Gain[V / V]times BW[GHz])/(P_{DC}[mW]times(F-1))$, this LNA achieves a competitive FOM among reported mm-Wave LNAs in literature [1–7].
{"title":"A Broadband Millimeter-Wave 5G Low Noise Amplifier Design in 22 nm Fully-Depleted Silicon-on-Insulator (FD-SOI) CMOS","authors":"Liang-Wei Ouyang, J. Mayeda, C. Sweeney, G. Somasundaram, D. Lie, Jerry Lopez","doi":"10.1109/IS3C57901.2023.00067","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00067","url":null,"abstract":"This paper presents a design of a broadband millimeter-wave (mm-Wave) low noise amplifier (LNA) designed in a 22 nm fully-depleted silicon-on-insulator (FD-SOI) CMOS technology. The post-layout parasitic extracted (PEX) simulations suggest the LNA has a 3-dB bandwidth (BW) from 16.9 – 41.8 GHz and a fractional bandwidth (FBW) of 84.8 %, covering the key frequency bands within the mm-Wave 5G FR2 band, with its noise figure (NF) ranging from 2.9 – 4.1 dB, and its input-referred 1-dB compression point (IP1dB) of −19.4 dBm at 28 GHz with 15.8 mW DC power consumption. Using the FOM (figure-of-merit) developed from Ref. [1] for broadband LNAs (FOM $equivmathbf{2 0}timeslog((Gain[V / V]times BW[GHz])/(P_{DC}[mW]times(F-1))$, this LNA achieves a competitive FOM among reported mm-Wave LNAs in literature [1–7].","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"20 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":"114466466","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/IS3C57901.2023.00073
Yau Yeu-Torng, Luu Thanh-Phu
In this paper, a dual-channel converter with a positive output and negative voltage output is proposed. It integrates a positive voltage output converter and a negative voltage output converter, and shares the same switches. The number of active components can be reduced. In addition, the circuit can achieve dual output voltage control with a single controller and PWM drive signal by appropriately designing the ratio of the number of windings of the coupling inductor. A regulated positive voltage output and negative voltage output can be achieved.
{"title":"A Dual-channel Converter with a Positive Output and a Negative Voltage Output","authors":"Yau Yeu-Torng, Luu Thanh-Phu","doi":"10.1109/IS3C57901.2023.00073","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00073","url":null,"abstract":"In this paper, a dual-channel converter with a positive output and negative voltage output is proposed. It integrates a positive voltage output converter and a negative voltage output converter, and shares the same switches. The number of active components can be reduced. In addition, the circuit can achieve dual output voltage control with a single controller and PWM drive signal by appropriately designing the ratio of the number of windings of the coupling inductor. A regulated positive voltage output and negative voltage output can be achieved.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"77 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":"125130683","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/IS3C57901.2023.00010
Hsieh Tung-Hsien, Jywe Wen-Yuh, Lai Hsin-Yu, Yi-Hao Chou, Wu Tsai-Hsu
The thermal error of machine tools is a key factor which affects machining accuracy. Currently, most inspection methods build a set of 3-axis or 5-axis non-contact measurement system using capacitance probes. However, since the equipment is expensive and not easy to set up, most thermal error model of machine tools can only be modeled beforehand. Therefore, once the AI model fails, it is often impossible to repair, or the equipment may be required to be brought to the manufacturing site again for installation, set-up, data collection and model building. In view of this, the study uses an optical non-contact spindle temperature measurement system previously developed by the team, which includes a 3D position sensing module, a standard glass ball (mounted on a standard tool holder interface), a PT100 temperature sensing module, an edge computer, and a human-machine interface. During the verification process, the system can effectively collect machine tool thermal data, including XYZ displacements, spindle speed, temperature, etc. By designing a quick tool holder jig, the center of the standard glass ball can be placed at the center of the 3D position sensor, significantly reducing the setup time. As for model building, this study uses XGBoost to establish correlation between temperature parameters and displacement in order to perform preliminary sensor selection. The RMSE and MSE of remaining sensors were then compared. After sensor selection, this study reduces the number of sensors used to 5, 7, 10, and 14. Then, LSTM and TCN is applied to build the thermal error model, with data from Day-1 (2022/07/15) as the training dataset. Using software and hardware modules mentioned in the study, thermal error for the test datasets Day-2 (2022/07/17) and Day-3 (2022/08/15) were decreased by more than 70%, which is also applicable to other dates.
{"title":"Development of LSTM and TCN Spindle Thermal Compensation Model by Using the Laser R-Test System","authors":"Hsieh Tung-Hsien, Jywe Wen-Yuh, Lai Hsin-Yu, Yi-Hao Chou, Wu Tsai-Hsu","doi":"10.1109/IS3C57901.2023.00010","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00010","url":null,"abstract":"The thermal error of machine tools is a key factor which affects machining accuracy. Currently, most inspection methods build a set of 3-axis or 5-axis non-contact measurement system using capacitance probes. However, since the equipment is expensive and not easy to set up, most thermal error model of machine tools can only be modeled beforehand. Therefore, once the AI model fails, it is often impossible to repair, or the equipment may be required to be brought to the manufacturing site again for installation, set-up, data collection and model building. In view of this, the study uses an optical non-contact spindle temperature measurement system previously developed by the team, which includes a 3D position sensing module, a standard glass ball (mounted on a standard tool holder interface), a PT100 temperature sensing module, an edge computer, and a human-machine interface. During the verification process, the system can effectively collect machine tool thermal data, including XYZ displacements, spindle speed, temperature, etc. By designing a quick tool holder jig, the center of the standard glass ball can be placed at the center of the 3D position sensor, significantly reducing the setup time. As for model building, this study uses XGBoost to establish correlation between temperature parameters and displacement in order to perform preliminary sensor selection. The RMSE and MSE of remaining sensors were then compared. After sensor selection, this study reduces the number of sensors used to 5, 7, 10, and 14. Then, LSTM and TCN is applied to build the thermal error model, with data from Day-1 (2022/07/15) as the training dataset. Using software and hardware modules mentioned in the study, thermal error for the test datasets Day-2 (2022/07/17) and Day-3 (2022/08/15) were decreased by more than 70%, which is also applicable to other dates.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"53 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":"125219834","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/IS3C57901.2023.00028
Hsin-Hui Wang, Chin-Yun Liu, S. Hung, Liang-Cheng Chen, Hui-Ling Hsieh, Wei-Min Liu
Radiotherapy is one of the common methods for cancer treatment. Developing a radiotherapy plan requires professional medical physicists or physicians to manually contour the organ boundaries in CT series, which is time-and labor-consuming. If artificial intelligence (AI) could assist with the task, it could alleviate the workload of medical staff, especially when medical resources are tight. We propose an AI-based automatic organ segmentation system trained by clinical datasets. However, this task is prone to be non-robust models in CT image series where the background occupies the majority of the scene. To remedy such data imbalance situation, we propose adopting three strategies during the model training steps: region classification, knowledge discovery in database, and sampler. The major segmentation task is based on U-Net and ResNet34 model where all convolution layers and batch normalization are replaced with group normalization and weight standardization to ensure effectiveness in small-batch data training. In this study, 33 organs throughout the body were segmented. The ablation experiments were conducted to prove all the training models have better performance than the original method. In the future, if a hospital needs to train model with their own private datasets, the three above strategies can be adopted to prevent unsuccessful training.
{"title":"Deep Learning Training Strategies for Severely Imbalanced Data in Organ Segmentation Tasks","authors":"Hsin-Hui Wang, Chin-Yun Liu, S. Hung, Liang-Cheng Chen, Hui-Ling Hsieh, Wei-Min Liu","doi":"10.1109/IS3C57901.2023.00028","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00028","url":null,"abstract":"Radiotherapy is one of the common methods for cancer treatment. Developing a radiotherapy plan requires professional medical physicists or physicians to manually contour the organ boundaries in CT series, which is time-and labor-consuming. If artificial intelligence (AI) could assist with the task, it could alleviate the workload of medical staff, especially when medical resources are tight. We propose an AI-based automatic organ segmentation system trained by clinical datasets. However, this task is prone to be non-robust models in CT image series where the background occupies the majority of the scene. To remedy such data imbalance situation, we propose adopting three strategies during the model training steps: region classification, knowledge discovery in database, and sampler. The major segmentation task is based on U-Net and ResNet34 model where all convolution layers and batch normalization are replaced with group normalization and weight standardization to ensure effectiveness in small-batch data training. In this study, 33 organs throughout the body were segmented. The ablation experiments were conducted to prove all the training models have better performance than the original method. In the future, if a hospital needs to train model with their own private datasets, the three above strategies can be adopted to prevent unsuccessful training.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"27 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":"115466222","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}
An intelligent gait parameter analysis system is proposed based on deep learning and human skeleton detection in videos. Video of the subject’s whole body while walking along a straight path is recorded, then gait landmark sequences are detected and corrected. After that, the corresponding frame intervals of heel landing are detected and used for calculating four gait parameters, gait speed, stride length, stride duration, and cadence. Experimental results have shown that by comparing each detected gait parameter with its corresponding ground truth, the mean squared error, mean absolute error, and mean absolute percentage error are all small. Moreover, five of six detected gait parameters possess high Pearson correlation coefficients with the corresponding ground truth. Therefore, our proposed system possesses the potential to be a precise and efficient gait analysis approach.
{"title":"Intelligent Gait Parameter Analysis System Based on Deep Learning and Human Skeleton Detection in Videos","authors":"Yi-Hung Chiu, Cheng-Yeh Tsai, Chen-Sen Ouyang, Chi-Hsien Huang, Yu-Chang Chen, San-Yuan Wang, Huei-Ping Dong","doi":"10.1109/IS3C57901.2023.00030","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00030","url":null,"abstract":"An intelligent gait parameter analysis system is proposed based on deep learning and human skeleton detection in videos. Video of the subject’s whole body while walking along a straight path is recorded, then gait landmark sequences are detected and corrected. After that, the corresponding frame intervals of heel landing are detected and used for calculating four gait parameters, gait speed, stride length, stride duration, and cadence. Experimental results have shown that by comparing each detected gait parameter with its corresponding ground truth, the mean squared error, mean absolute error, and mean absolute percentage error are all small. Moreover, five of six detected gait parameters possess high Pearson correlation coefficients with the corresponding ground truth. Therefore, our proposed system possesses the potential to be a precise and efficient gait analysis approach.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"15 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":"129627717","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/IS3C57901.2023.00038
Pai-Hsun Chen, Lu-Han Chen, Yin-Nan Wang
Virtual reality (VR) technology has advanced significantly in recent years. However, effective VR scene design remains challenging as it is both laborious and time-consuming. Although studies on immersive 3D scene editing have been conducted, it is currently limited to laboratory-specific designs or professional VR equipment. This study aims to develop immersive authoring systems that provide portability, accessibility, reusability, sustainability, cost-effectiveness, epidemic safety, mobility, immersion, and intuition not present in current commercial or lab-specific VR. To develop such a system, it is necessary to consider both the physical and software designs of the interaction, as VR-TUI is not only a computer device and software program but also a headset and an interactive physical controller. To satisfy the said research goal, the researchers of this study proposed an approach that integrates a rational unified process with design thinking to address the entire VR-TUI development and design. This approach can be used to develop human-centered products or systems that require integrating various software designs, hardware designs, physical designs, and other product or system design areas, providing an alternative reference development process and methodology for cross-disciplinary technology product development.
{"title":"Integrating Rational Unified Process and Design Thinking Approach to Develop A Tangible User Interface for Mobile VR Immersive Scenarios Authoring Prototype","authors":"Pai-Hsun Chen, Lu-Han Chen, Yin-Nan Wang","doi":"10.1109/IS3C57901.2023.00038","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00038","url":null,"abstract":"Virtual reality (VR) technology has advanced significantly in recent years. However, effective VR scene design remains challenging as it is both laborious and time-consuming. Although studies on immersive 3D scene editing have been conducted, it is currently limited to laboratory-specific designs or professional VR equipment. This study aims to develop immersive authoring systems that provide portability, accessibility, reusability, sustainability, cost-effectiveness, epidemic safety, mobility, immersion, and intuition not present in current commercial or lab-specific VR. To develop such a system, it is necessary to consider both the physical and software designs of the interaction, as VR-TUI is not only a computer device and software program but also a headset and an interactive physical controller. To satisfy the said research goal, the researchers of this study proposed an approach that integrates a rational unified process with design thinking to address the entire VR-TUI development and design. This approach can be used to develop human-centered products or systems that require integrating various software designs, hardware designs, physical designs, and other product or system design areas, providing an alternative reference development process and methodology for cross-disciplinary technology product development.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"1 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":"131265474","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/IS3C57901.2023.00035
Yu-Chia Kao, Wei-Hsuan Chen, S. Ueng
In this article, we propose an innovative algorithm for texture-mapping voxel-based models. Voxel-based models are composed of voxels. Their surfaces are digitalized and basic geometrical information, like normal and tangent vectors, are absent from their representations. Relying on connectivity and geometrical information to parametrize the surface of a voxel-based model is impossible. Instead, we derive an automatic mapping procedure, based on Self-Organizing Map (SOM), to parametrize its surface voxels. First, we use an unsupervised training to convert the SOM lattice into an approximation surface of the model by using the surface voxels as input data. Then, another unsupervised training is triggered to parameterize the nodes of the SOM lattice by using the texels of the texture as input data. In the $3^{rd}$ stage, the surface voxels are textured, based on the relations established in the two training processes. As a result, the mapping task is efficiently accomplished without too much human interference.
{"title":"Texture Mapping for Voxel Models Using SOM","authors":"Yu-Chia Kao, Wei-Hsuan Chen, S. Ueng","doi":"10.1109/IS3C57901.2023.00035","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00035","url":null,"abstract":"In this article, we propose an innovative algorithm for texture-mapping voxel-based models. Voxel-based models are composed of voxels. Their surfaces are digitalized and basic geometrical information, like normal and tangent vectors, are absent from their representations. Relying on connectivity and geometrical information to parametrize the surface of a voxel-based model is impossible. Instead, we derive an automatic mapping procedure, based on Self-Organizing Map (SOM), to parametrize its surface voxels. First, we use an unsupervised training to convert the SOM lattice into an approximation surface of the model by using the surface voxels as input data. Then, another unsupervised training is triggered to parameterize the nodes of the SOM lattice by using the texels of the texture as input data. In the $3^{rd}$ stage, the surface voxels are textured, based on the relations established in the two training processes. As a result, the mapping task is efficiently accomplished without too much human interference.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"10 23 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":"131266485","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/IS3C57901.2023.00061
M. Ida
In this paper we examine education related data of higher education institutions or universities in Japan. Especially we examine eigenvalues of correlation analysis for universities’ data of student mobility by using the knowledge of Random Matrix Theory. We show some numerical examples to examine the effectiveness of the knowledge for eigenvalues and its application to Principal Component Analysis. Moreover, we identify the future issues of this analysis method.
{"title":"Eigenvalues of Correlation Analysis for Higher Education Institutional Data","authors":"M. Ida","doi":"10.1109/IS3C57901.2023.00061","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00061","url":null,"abstract":"In this paper we examine education related data of higher education institutions or universities in Japan. Especially we examine eigenvalues of correlation analysis for universities’ data of student mobility by using the knowledge of Random Matrix Theory. We show some numerical examples to examine the effectiveness of the knowledge for eigenvalues and its application to Principal Component Analysis. Moreover, we identify the future issues of this analysis method.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"1 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":"130589646","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/IS3C57901.2023.00095
Wei-Jong Yang, P. Chung
Federated learning provides a decentralized learning without data exchange. Among them, the Federated Average (FedAVG) framework is the most likely to be implemented in real world application due to its low communication overhead. However, this architecture can easily affect the efficiency of global model convergence when there are differences data distribution in individual user. Therefore, in this paper, we propose an aggregation strategy called significant Weighted feature aggregation method, in which the features with large variation are appropriately weighted at the server side to improve the model convergence speed even in not identically and independently distributed (non-iid) environments. As shown in our experiments, our approach had over 10% of improvements compared to the FedAVG.
{"title":"Significant Weighted Aggregation Method for Federated Learning in Non-iid Environment","authors":"Wei-Jong Yang, P. Chung","doi":"10.1109/IS3C57901.2023.00095","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00095","url":null,"abstract":"Federated learning provides a decentralized learning without data exchange. Among them, the Federated Average (FedAVG) framework is the most likely to be implemented in real world application due to its low communication overhead. However, this architecture can easily affect the efficiency of global model convergence when there are differences data distribution in individual user. Therefore, in this paper, we propose an aggregation strategy called significant Weighted feature aggregation method, in which the features with large variation are appropriately weighted at the server side to improve the model convergence speed even in not identically and independently distributed (non-iid) environments. As shown in our experiments, our approach had over 10% of improvements compared to the FedAVG.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"1 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":"130242378","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/IS3C57901.2023.00022
Jui-Chi Chen, Zhen-You Lian, Hsin-You Chiang, Chung-Lin Huang, C. Chuang
In recent years, there has been a rapid development of intelligent driving assistance systems. Although most vehicles nowadays are equipped with driving assistance systems, the number of car accidents continues to rise. The main cause of car accidents is still largely attributed to human factors. Therefore, there has been an increasing focus on research related to accident detection and driver behavior analysis. This study used deep learning methods to automatically recognize driving events from recorded driving videos. In the training phase of deep learning, we cropped all the videos in the training data into multiple clips, and labeled driving event categories for each clip, including four categories: vehicle stopped, straight driving, turning, and collision. The proposed model references the architecture of the SlowFastNet model and the concepts of I3D. We expanded Inception-V3 to a 3D structure and replaced the bottom architecture of SlowFastNet with 3D-Inception-V3, making the network more applicable to the training data. After training, the model can recognize driving events in various driving environments. Through experimental comparisons, our network architecture achieved the highest recognition accuracy, with an accuracy rate of 93.3%.
{"title":"Automatic Recognition of Driving Events based on Deep Learning","authors":"Jui-Chi Chen, Zhen-You Lian, Hsin-You Chiang, Chung-Lin Huang, C. Chuang","doi":"10.1109/IS3C57901.2023.00022","DOIUrl":"https://doi.org/10.1109/IS3C57901.2023.00022","url":null,"abstract":"In recent years, there has been a rapid development of intelligent driving assistance systems. Although most vehicles nowadays are equipped with driving assistance systems, the number of car accidents continues to rise. The main cause of car accidents is still largely attributed to human factors. Therefore, there has been an increasing focus on research related to accident detection and driver behavior analysis. This study used deep learning methods to automatically recognize driving events from recorded driving videos. In the training phase of deep learning, we cropped all the videos in the training data into multiple clips, and labeled driving event categories for each clip, including four categories: vehicle stopped, straight driving, turning, and collision. The proposed model references the architecture of the SlowFastNet model and the concepts of I3D. We expanded Inception-V3 to a 3D structure and replaced the bottom architecture of SlowFastNet with 3D-Inception-V3, making the network more applicable to the training data. After training, the model can recognize driving events in various driving environments. Through experimental comparisons, our network architecture achieved the highest recognition accuracy, with an accuracy rate of 93.3%.","PeriodicalId":142483,"journal":{"name":"2023 Sixth International Symposium on Computer, Consumer and Control (IS3C)","volume":"1 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":"130728323","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}