Pub Date : 2025-03-01DOI: 10.23919/cje.2023.00.257
Baodi Liu;Jing Tian;Zhenlong Wang;Weifeng Liu;Xinan Yuan;Wei Li
Most existing deep learning-based underwater image enhancement methods rely heavily on synthetic paired underwater images, which limits their practicality and generalization. Unsupervised underwater image enhancement methods can be trained on unpaired data, overcoming the reliance on paired data. However, existing unsupervised methods suffer from poor color correction capability, artifacts, and blurry details in the generated images. Therefore, this paper proposes a dual generative adversarial network (GAN) with contrastive learning constraints to achieve unsupervised underwater image enhancement. Firstly, we construct a dual GAN network for image transformation. Secondly, we utilize patch-based learning to maximize the mutual information between inputs and outputs, eliminating the reliance on paired data. Thirdly, we use image gradient difference loss to mitigate artifacts in the generated images. Lastly, to address the problem of blurry details, we incorporate channel attention in the generator network to focus on more important content and improve the quality of the generated images. Extensive experiments demonstrate that the enhanced results of our method show amelioration in visual quality.
{"title":"DCUGAN: Dual Contrastive Learning GAN for Unsupervised Underwater Image Enhancement","authors":"Baodi Liu;Jing Tian;Zhenlong Wang;Weifeng Liu;Xinan Yuan;Wei Li","doi":"10.23919/cje.2023.00.257","DOIUrl":"https://doi.org/10.23919/cje.2023.00.257","url":null,"abstract":"Most existing deep learning-based underwater image enhancement methods rely heavily on synthetic paired underwater images, which limits their practicality and generalization. Unsupervised underwater image enhancement methods can be trained on unpaired data, overcoming the reliance on paired data. However, existing unsupervised methods suffer from poor color correction capability, artifacts, and blurry details in the generated images. Therefore, this paper proposes a dual generative adversarial network (GAN) with contrastive learning constraints to achieve unsupervised underwater image enhancement. Firstly, we construct a dual GAN network for image transformation. Secondly, we utilize patch-based learning to maximize the mutual information between inputs and outputs, eliminating the reliance on paired data. Thirdly, we use image gradient difference loss to mitigate artifacts in the generated images. Lastly, to address the problem of blurry details, we incorporate channel attention in the generator network to focus on more important content and improve the quality of the generated images. Extensive experiments demonstrate that the enhanced results of our method show amelioration in visual quality.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 3","pages":"906-916"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11060021","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144519349","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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}
This article describes the result of work by the GCC (GNU compiler collection) compiler team to improve the performance of CPUBench tests on the Kunpeng 920 platform. During performance analysis, certain deficiencies were discovered, which were eliminated by modifying the GCC compiler. Overall, around 10 optimizations were introduced to openEuler GCC. Some of them improve the existing optimizations, while others are independent optimization passes. The result of the work was an improvement in the performance of the CPUBench integer test package by more than 12% on single-core run and more than 11% on multi-core run, with an improvement of individual tests up to 74%, as well as an improvement of the SPEC CPU 2017 integer package by around 1.4%.
本文描述了GCC （GNU编译器集合）编译器团队在鲲鹏920平台上改进CPUBench测试性能的工作结果。在性能分析期间，发现了某些缺陷，通过修改GCC编译器可以消除这些缺陷。总的来说，openEuler GCC中引入了大约10个优化。其中一些改进了现有的优化，而另一些则是独立的优化过程。这项工作的结果是，CPUBench整数测试包在单核运行时的性能提高了12%以上，在多核运行时的性能提高了11%以上，单个测试的性能提高了74%，SPEC CPU 2017整数包的性能提高了1.4%左右。
{"title":"Development of GCC Optimizations to Speed up CPUBench Integer Benchmarks on ARMv8.2","authors":"Viacheslav Chernonog;Andrey Dobrov;Ilia Diachkov;Alexander Pronin;Egor Melnichenko;Emin Gadzhiev","doi":"10.23919/cje.2024.00.105","DOIUrl":"https://doi.org/10.23919/cje.2024.00.105","url":null,"abstract":"This article describes the result of work by the GCC (GNU compiler collection) compiler team to improve the performance of CPUBench tests on the Kunpeng 920 platform. During performance analysis, certain deficiencies were discovered, which were eliminated by modifying the GCC compiler. Overall, around 10 optimizations were introduced to openEuler GCC. Some of them improve the existing optimizations, while others are independent optimization passes. The result of the work was an improvement in the performance of the CPUBench integer test package by more than 12% on single-core run and more than 11% on multi-core run, with an improvement of individual tests up to 74%, as well as an improvement of the SPEC CPU 2017 integer package by around 1.4%.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 3","pages":"962-969"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11060052","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144519308","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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 : 2025-03-01DOI: 10.23919/cje.2024.00.070
Tianhao Fu;Zehua Yang;Zhisheng Ye;Chenxiang Ma;Yang Han;Yingwei Luo;Xiaolin Wang;Zhenlin Wang
As deep learning (DL) technology rapidly advances in areas such as computer vision, natural language processing, and more recently, large language models (LLMs), the demand for computing resources has increasingly grown. In particular, scheduling deep learning training (DLT) jobs on graphics processing unit (GPU) clusters has become crucial for the effective utilization of computing resources and the acceleration of model training processes. However, resource management and scheduling in GPU clusters face challenges related to computing and communication, including job sharing, interference, elastic scheduling, heterogeneous resources, and fairness. This survey investigates the scheduling issues of DLT jobs in GPU clusters, focusing on scheduling optimizations at the job characteristic and cluster resource levels. We analyze the structure and training computing characteristics of traditional DL models and LLMs, as well as their requirements for iterative computation, communication, GPU sharing, and resource elasticity. In addition, we compare the main contributions of this survey with related reviews and discuss research directions, including scheduling based on job characteristics and optimization strategies for cluster resources. This survey aims to provide researchers and practitioners with a comprehensive understanding of DLT job scheduling in GPU clusters and to point out directions for future research.
{"title":"A Survey on the Scheduling of DL and LLM Training Jobs in GPU Clusters","authors":"Tianhao Fu;Zehua Yang;Zhisheng Ye;Chenxiang Ma;Yang Han;Yingwei Luo;Xiaolin Wang;Zhenlin Wang","doi":"10.23919/cje.2024.00.070","DOIUrl":"https://doi.org/10.23919/cje.2024.00.070","url":null,"abstract":"As deep learning (DL) technology rapidly advances in areas such as computer vision, natural language processing, and more recently, large language models (LLMs), the demand for computing resources has increasingly grown. In particular, scheduling deep learning training (DLT) jobs on graphics processing unit (GPU) clusters has become crucial for the effective utilization of computing resources and the acceleration of model training processes. However, resource management and scheduling in GPU clusters face challenges related to computing and communication, including job sharing, interference, elastic scheduling, heterogeneous resources, and fairness. This survey investigates the scheduling issues of DLT jobs in GPU clusters, focusing on scheduling optimizations at the job characteristic and cluster resource levels. We analyze the structure and training computing characteristics of traditional DL models and LLMs, as well as their requirements for iterative computation, communication, GPU sharing, and resource elasticity. In addition, we compare the main contributions of this survey with related reviews and discuss research directions, including scheduling based on job characteristics and optimization strategies for cluster resources. This survey aims to provide researchers and practitioners with a comprehensive understanding of DLT job scheduling in GPU clusters and to point out directions for future research.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 3","pages":"881-905"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11060018","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144519310","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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 : 2025-03-01DOI: 10.23919/cje.2024.00.224
Lin Gu;Yuandan Dong
A novel varactor-loaded microstrip resonator with a hybrid structure of microstrip and parallel-coupled lines has been proposed. Stopbands are constructed utilizing the first-and second-mode of the resonator, respectively. The independent control of the two modes of this resonator was theoretically analyzed and validated. The introduction of parallel-coupled lines was employed to enhance stopband attenuation, mitigating the impact of low- $Q$ value varactor diodes to some extent. Miniaturization is also achieved through the introduction of two sections of parallel-coupled lines. A 2nd-order dual-band tunable bandstop filter was designed, fabricated, and measured, with the measured results revealing high attenuation levels of 36.4 dB and 27.85 dB for the two stopbands, respectively, in addition to a compact size of $0.23lambda_{g}times 0.07lambda_{g}$ (where $lambda_{g}$ is the guided wavelength in the substrate at 2.2 GHz).
{"title":"Miniaturized Reconfigurable Dual-Band Bandstop Filter Utilizing a Novel Hybrid Resonator for Enhanced Stopband Suppression","authors":"Lin Gu;Yuandan Dong","doi":"10.23919/cje.2024.00.224","DOIUrl":"https://doi.org/10.23919/cje.2024.00.224","url":null,"abstract":"A novel varactor-loaded microstrip resonator with a hybrid structure of microstrip and parallel-coupled lines has been proposed. Stopbands are constructed utilizing the first-and second-mode of the resonator, respectively. The independent control of the two modes of this resonator was theoretically analyzed and validated. The introduction of parallel-coupled lines was employed to enhance stopband attenuation, mitigating the impact of low- <tex>$Q$</tex> value varactor diodes to some extent. Miniaturization is also achieved through the introduction of two sections of parallel-coupled lines. A 2nd-order dual-band tunable bandstop filter was designed, fabricated, and measured, with the measured results revealing high attenuation levels of 36.4 dB and 27.85 dB for the two stopbands, respectively, in addition to a compact size of <tex>$0.23lambda_{g}times 0.07lambda_{g}$</tex> (where <tex>$lambda_{g}$</tex> is the guided wavelength in the substrate at 2.2 GHz).","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 3","pages":"766-773"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11060046","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144519311","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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 : 2025-03-01DOI: 10.23919/cje.2023.00.280
Jian Su;Shiang Mao;Wei Zhuang
In the context of complex foggy environments, the acquired images often suffer from low visibility, high noise, and loss of detailed information. The direct application of general object detection methods fails to achieve satisfactory results. To address these issues, this paper proposes a foggy object detection method based on YOLOv8n, named AOYOLO. The all-in-one dehazing network, a lightweight defogging network, is employed for data augmentation. Additionally, the ResCNet module is introduced in the backbone to better extract features from low-illumination images. The GACSP module is proposed in the neck to capture multi-scale features and effectively utilize them, thereby generating discriminative features with different scales. The detection head is improved using WiseIoU, which enhances the accuracy of object localization. Experimental evaluations are conducted on the publicly available datasets: the annotated real-world task-driven testing set (RTTS) and synthetic foggy KITTI dataset. The results demonstrate that the proposed AOYOLO algorithm outperforms the original YOLOv8n algorithm with an average mean average precision (mAP) improvement of 3.3% and 4.6% on the RTTS and KITTI datasets, respectively. The AOYOLO method effectively enhances the performance of object detection in foggy scenes. Due to its improved performance and stronger robustness, this experimental model provides a new perspective for foggy object detection.
{"title":"AOYOLO Algorithm Oriented Vehicle and Pedestrian Detection in Foggy Weather","authors":"Jian Su;Shiang Mao;Wei Zhuang","doi":"10.23919/cje.2023.00.280","DOIUrl":"https://doi.org/10.23919/cje.2023.00.280","url":null,"abstract":"In the context of complex foggy environments, the acquired images often suffer from low visibility, high noise, and loss of detailed information. The direct application of general object detection methods fails to achieve satisfactory results. To address these issues, this paper proposes a foggy object detection method based on YOLOv8n, named AOYOLO. The all-in-one dehazing network, a lightweight defogging network, is employed for data augmentation. Additionally, the ResCNet module is introduced in the backbone to better extract features from low-illumination images. The GACSP module is proposed in the neck to capture multi-scale features and effectively utilize them, thereby generating discriminative features with different scales. The detection head is improved using WiseIoU, which enhances the accuracy of object localization. Experimental evaluations are conducted on the publicly available datasets: the annotated real-world task-driven testing set (RTTS) and synthetic foggy KITTI dataset. The results demonstrate that the proposed AOYOLO algorithm outperforms the original YOLOv8n algorithm with an average mean average precision (mAP) improvement of 3.3% and 4.6% on the RTTS and KITTI datasets, respectively. The AOYOLO method effectively enhances the performance of object detection in foggy scenes. Due to its improved performance and stronger robustness, this experimental model provides a new perspective for foggy object detection.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 2","pages":"661-672"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10982094","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143900521","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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}
This article presents a new method for substrate integrated waveguide (SIW) filters to achieve wide stopbands. Using the proposed staggered inter-coupling structures, double-layer SIW filters working at the fundamental mode TE101(f0) can have wide stopbands up to TE10(2n-1), where $n$ is the order of the filter. They can break the upper limit of the stopband extension and have coplanar ports suitable for planar circuits and systems in comparison to their multilayer counterparts, and they can further extend the stopbands and have shielding structures suitable for high-performance and high-frequency applications compared to their hybrid counterparts. Three examples are provided. The measured results show that they respectively achieve wide stopbands up to 3.97f0, 5.22f0, and 6.53f0. The proposed technique should be effective for developing wide stopband SIW filters for microwave circuits and systems.
{"title":"TE101 Substrate Integrated Waveguide Filter with Wide Stopband Up to TE10(2n-l) and Coplanar Ports","authors":"Peng Chu;Jianguo Feng;Lei Guo;Fang Zhu;Weibin Kong;Leilei Liu;Guoqing Luo;Ke Wu","doi":"10.23919/cje.2023.00.225","DOIUrl":"https://doi.org/10.23919/cje.2023.00.225","url":null,"abstract":"This article presents a new method for substrate integrated waveguide (SIW) filters to achieve wide stopbands. Using the proposed staggered inter-coupling structures, double-layer SIW filters working at the fundamental mode TE<inf>101</inf>(f<inf>0</inf>) can have wide stopbands up to TE<inf>10(2n-1)</inf>, where <tex>$n$</tex> is the order of the filter. They can break the upper limit of the stopband extension and have coplanar ports suitable for planar circuits and systems in comparison to their multilayer counterparts, and they can further extend the stopbands and have shielding structures suitable for high-performance and high-frequency applications compared to their hybrid counterparts. Three examples are provided. The measured results show that they respectively achieve wide stopbands up to 3.97f<inf>0</inf>, 5.22f<inf>0</inf>, and 6.53f<inf>0</inf>. The proposed technique should be effective for developing wide stopband SIW filters for microwave circuits and systems.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 2","pages":"457-463"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10982088","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143900570","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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 : 2025-03-01DOI: 10.23919/cje.2023.00.061
Genggeng Liu;Ling Wei;Yantao Yu;Ning Xu
As advanced technology nodes enter the nanometer era, the complexity of integrated circuit design is increasing, and the proportion of bus in the net is also increasing. The bus routing has become a key factor affecting the performance of the chip. In addition, the existing research does not distinguish between bus and non-bus in the complete global routing process, which directly leads to the expansion of bus deviation and the degradation of chip performance. In order to solve these problems, we propose a high-quality and efficient bus-aware global router, which includes the following key strategies: By introducing the routing density graph, we propose a routing model that can simultaneously consider the routability of non-bus and the deviation value of bus; A dynamic routing resource adjustment algorithm is proposed to optimize the bus deviation and wirelength simultaneously, which can effectively reduce the bus deviation; We propose a layer assignment algorithm consider deviation to significantly reduce the bus deviation of the 3D routing solution; And a depth-first search (DFS)-based algorithm is proposed to obtain multiple routing solutions, from which the routing result with the lowest deviation is selected. Experimental results show that the proposed algorithms can effectively reduce bus deviation compared with the existing algorithms, so as to obtain high-quality 2D and 3D routing solutions considering bus deviation.
{"title":"A High-Quality and Efficient Bus-Aware Global Router","authors":"Genggeng Liu;Ling Wei;Yantao Yu;Ning Xu","doi":"10.23919/cje.2023.00.061","DOIUrl":"https://doi.org/10.23919/cje.2023.00.061","url":null,"abstract":"As advanced technology nodes enter the nanometer era, the complexity of integrated circuit design is increasing, and the proportion of bus in the net is also increasing. The bus routing has become a key factor affecting the performance of the chip. In addition, the existing research does not distinguish between bus and non-bus in the complete global routing process, which directly leads to the expansion of bus deviation and the degradation of chip performance. In order to solve these problems, we propose a high-quality and efficient bus-aware global router, which includes the following key strategies: By introducing the routing density graph, we propose a routing model that can simultaneously consider the routability of non-bus and the deviation value of bus; A dynamic routing resource adjustment algorithm is proposed to optimize the bus deviation and wirelength simultaneously, which can effectively reduce the bus deviation; We propose a layer assignment algorithm consider deviation to significantly reduce the bus deviation of the 3D routing solution; And a depth-first search (DFS)-based algorithm is proposed to obtain multiple routing solutions, from which the routing result with the lowest deviation is selected. Experimental results show that the proposed algorithms can effectively reduce bus deviation compared with the existing algorithms, so as to obtain high-quality 2D and 3D routing solutions considering bus deviation.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 2","pages":"444-456"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10982079","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143900590","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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 : 2025-03-01DOI: 10.23919/cje.2023.00.178
Yi Zhang;Yuhang Zhuang;Hu Zhang;Lei Yang;Jing Wang;Changchun Zhang;Yufeng Guo
In this study, a broadband monolithic microwave integrated circuit (MMIC) double-balanced mixer designed for operation within the frequency range of 8–26 GHz is presented. The design is implemented using a 0.15 μm GaAs process. Traditional Marchand baluns, when applied to wideband mixers, face challenges in simultaneously achieving broad bandwidth and good port matching characteristics. To address this issue, we employ a spiral Marchand balun with a compensation capacitor. This innovative approach not only maintains the mixer's wide bandwidth but also enhances the matching between the local oscillator (LO) and radio frequency (RF) ports. Additionally, it significantly simplifies the complexity of designing the matching circuit. The optimization principle of the compensation capacitor is elaborated in detail within this paper. Experimental results demonstrate that, with an LO power of 14 dBm, the conversion loss remains below 8.5 dB, while the voltage standing wave ratio (VSWR) of the LO and IF ports is less than 2 and the VSWR of the RF port is below 2.4. In comparison with existing literature, our designed mixer exhibits a broader bandwidth and lower loss.
{"title":"An 8–26 GHz Passive Mixer with Excellent Port Matching Utilizing Marchand Balun and Capacitor Compensation","authors":"Yi Zhang;Yuhang Zhuang;Hu Zhang;Lei Yang;Jing Wang;Changchun Zhang;Yufeng Guo","doi":"10.23919/cje.2023.00.178","DOIUrl":"https://doi.org/10.23919/cje.2023.00.178","url":null,"abstract":"In this study, a broadband monolithic microwave integrated circuit (MMIC) double-balanced mixer designed for operation within the frequency range of 8–26 GHz is presented. The design is implemented using a 0.15 μm GaAs process. Traditional Marchand baluns, when applied to wideband mixers, face challenges in simultaneously achieving broad bandwidth and good port matching characteristics. To address this issue, we employ a spiral Marchand balun with a compensation capacitor. This innovative approach not only maintains the mixer's wide bandwidth but also enhances the matching between the local oscillator (LO) and radio frequency (RF) ports. Additionally, it significantly simplifies the complexity of designing the matching circuit. The optimization principle of the compensation capacitor is elaborated in detail within this paper. Experimental results demonstrate that, with an LO power of 14 dBm, the conversion loss remains below 8.5 dB, while the voltage standing wave ratio (VSWR) of the LO and IF ports is less than 2 and the VSWR of the RF port is below 2.4. In comparison with existing literature, our designed mixer exhibits a broader bandwidth and lower loss.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 2","pages":"422-428"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10982081","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143900628","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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 Gaussian mixture model (GMM) is a classical probabilistic representation model widely used in unsupervised learning. GMM performs poorly on high-dimensional data (HDD) due to the requirement of estimating a large number of parameters with relatively few observations. To address this, the paper proposes a novel subspace-based GMM clustering ensemble (SubGMM-CE) algorithm tailored for HDD. The proposed SubGMM-CE algorithm comprises three key components. A series of low-dimensional subspaces are dynamically determined, considering the optimal number of GMM components. The GMM-based clustering algorithm is applied to each subspace to obtain a series of heterogeneous GMM models. These GMM base clustering results are merged using the newly-designed relabeling strategy based on the average shared affiliation probability, generating the final clustering result for high-dimensional unlabeled data. An exhaustive experimental evaluation validates the feasibility, rationality, effectiveness, and robustness to noise of the SubGMM-CE algorithm. Results show that SubGMM-CE achieves higher stability and more accurate clustering results, outperforming nine state-of-the-art clustering algorithms in normalized mutual information, clustering accuracy, and adjusted rand index scores. This demonstrates the viability of the SubGMM-CE algorithm in addressing HDD clustering challenges.
{"title":"A Novel Subspace-Based GMM Clustering Ensemble Algorithm for High-Dimensional Data","authors":"Yulin He;Yingting He;Zhaowu Zhan;Fournier-Viger Philippe;Joshua Zhexue Huang","doi":"10.23919/cje.2023.00.153","DOIUrl":"https://doi.org/10.23919/cje.2023.00.153","url":null,"abstract":"The Gaussian mixture model (GMM) is a classical probabilistic representation model widely used in unsupervised learning. GMM performs poorly on high-dimensional data (HDD) due to the requirement of estimating a large number of parameters with relatively few observations. To address this, the paper proposes a novel subspace-based GMM clustering ensemble (SubGMM-CE) algorithm tailored for HDD. The proposed SubGMM-CE algorithm comprises three key components. A series of low-dimensional subspaces are dynamically determined, considering the optimal number of GMM components. The GMM-based clustering algorithm is applied to each subspace to obtain a series of heterogeneous GMM models. These GMM base clustering results are merged using the newly-designed relabeling strategy based on the average shared affiliation probability, generating the final clustering result for high-dimensional unlabeled data. An exhaustive experimental evaluation validates the feasibility, rationality, effectiveness, and robustness to noise of the SubGMM-CE algorithm. Results show that SubGMM-CE achieves higher stability and more accurate clustering results, outperforming nine state-of-the-art clustering algorithms in normalized mutual information, clustering accuracy, and adjusted rand index scores. This demonstrates the viability of the SubGMM-CE algorithm in addressing HDD clustering challenges.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 2","pages":"612-629"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10982075","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143900635","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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 : 2025-03-01DOI: 10.23919/cje.2023.00.419
Hao Wu;Junqi Guo;Rongfang Bie
Dear Editor, Retrieving target images accurately shows more and more prominent significance in the era of digital media and big data. Although there are many classic methods proposed, the overwhelming majority of them are still improved based on the strategy of machine learning. In recent years, deep learning models (such as convolutional neural networks [1]–[3], restricted Boltzmann machines [4], [5], autoencoders [6]–[8], and sparse coding [9], [10]) have used more complicated networks to extract essential features more completely. Moreover, the overwhelming advantages of experimental results support it to replace the traditional machine learning methods in a short while. On the basis of classic models, many innovative models [11], [12] have been proposed, demonstrating better practical application value. Although we must admit that deep learning models have provided revolutionary changes, the huge computing resource consumption is also a burden that can not be underestimated. Even if some methods can reduce the amount of learning instances relatively, they are at the cost of accuracy reduction in most cases, and even some models have obvious limitations which are only effective for some categories.
{"title":"Multi-Use Learning Instance for Optimized Image Retrieval","authors":"Hao Wu;Junqi Guo;Rongfang Bie","doi":"10.23919/cje.2023.00.419","DOIUrl":"https://doi.org/10.23919/cje.2023.00.419","url":null,"abstract":"Dear Editor, Retrieving target images accurately shows more and more prominent significance in the era of digital media and big data. Although there are many classic methods proposed, the overwhelming majority of them are still improved based on the strategy of machine learning. In recent years, deep learning models (such as convolutional neural networks [1]–[3], restricted Boltzmann machines [4], [5], autoencoders [6]–[8], and sparse coding [9], [10]) have used more complicated networks to extract essential features more completely. Moreover, the overwhelming advantages of experimental results support it to replace the traditional machine learning methods in a short while. On the basis of classic models, many innovative models [11], [12] have been proposed, demonstrating better practical application value. Although we must admit that deep learning models have provided revolutionary changes, the huge computing resource consumption is also a burden that can not be underestimated. Even if some methods can reduce the amount of learning instances relatively, they are at the cost of accuracy reduction in most cases, and even some models have obvious limitations which are only effective for some categories.","PeriodicalId":50701,"journal":{"name":"Chinese Journal of Electronics","volume":"34 3","pages":"1002-1005"},"PeriodicalIF":1.6,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11060047","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144519339","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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：