Article
Computer Science, Information Systems
Chao Tong, Mengze Zhang, Chao Lang, Zhigao Zheng
Summary: A proposed intelligent algorithm aims to prevent deep neural network detectors from detecting personal private information, particularly human faces, while minimizing the impact on image quality. By training and generating adversarial samples, as well as improving the model, the method significantly interferes with DNN detectors with minimal impact on visual quality.
ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS
(2021)
Article
Energy & Fuels
Kuang-Yow Lian, Yong-Jie Hong, Che-Wei Chang, Yu-Wei Su
Summary: This paper proposes a new method called the backward modeling approach (BMA) to achieve optimal chiller loading (OCL) for reducing energy consumption in industries with multiple-chillers of different efficiency. The developed OCL regulator (OCLR) based on the novel BMA approach consists of a conditional generative network (cGAN) and a deep neural network (DNN). By using the developed OCLR, the chilled water supply temperature can be set to achieve the desired energy saving. The experimental results showed that the data-driven OCLR based on BMA has high performance and was able to save significant energy.
Article
Computer Science, Artificial Intelligence
Yuesong Tian, Li Shen, Guinan Su, Zhifeng Li, Wei Liu
Summary: This paper introduces a fully differentiable search framework called alphaGAN, which efficiently obtains suitable network architectures from a large search space, further advancing the learning and performance improvement of generative models. The experiments demonstrate satisfactory results on multiple datasets and provide a comprehensive analysis of the searching process and searched architectures.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Tongtong Gao, Yue Zhou, Shukai Duan, Xiaofang Hu
Summary: This paper presents a BNN training framework called KDG-BNN, which consists of a full-precision network, a 1-bit binary network, and a discriminator. The framework aims to bridge the performance gap between BNNs and FNNs through the use of distillation loss and adversarial training. Additionally, the paper proposes a memristor-based implementation scheme for KDG-BNN, leveraging the advantages of memristors and lightweight BNNs for intelligent application deployment.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Chemistry, Multidisciplinary
Hee-Deok Yang
Summary: The article discusses how to use an attentive generative adversarial network (ATTGAN) to remove image noise caused by raindrops and other inclement weather conditions, improving the performance of outdoor vision systems.
APPLIED SCIENCES-BASEL
(2021)
Article
Engineering, Electrical & Electronic
Wenbin Zhu, Hong Gu, Xiaochun Zhu
Summary: This paper proposes a dual-network classification model based on contrastive learning and pseudo-label training strategy to solve the classification task of an SAR image with limited data. It achieves higher classification accuracy through unsupervised pre-training and fine-tuning using pseudo-label information.
IET RADAR SONAR AND NAVIGATION
(2022)
Article
Green & Sustainable Science & Technology
Sertan Serte, Mehmet Alp Dirik, Fadi Al-Turjman
Summary: Healthcare is enhanced through the Internet of things, with machine learning-based systems providing faster services and doctors utilizing artificial intelligence to analyze X-rays and CT scans. This paper proposes a data-efficient deep network that generates synthetic CT scans using a generative adversarial network (GAN) to increase the available data. The GAN-based deep learning model shows superior performance in COVID-19 detection compared to classic models, as evaluated on the COVID19-CT and Mosmed datasets.
Article
Computer Science, Artificial Intelligence
Mengchen Shi, Fei Xie, Jiquan Yang, Jing Zhao, Xixiang Liu, Fan Wang
Summary: Generative adversarial networks benefit from large datasets and carefully chosen model parameters. Cutout with patch-loss augmentation is proposed as a solution to prevent overfitting and mode collapse. Additional technique of tensor value clamp is introduced to accelerate training speed without sacrificing quality. The method can be applied to different generative adversarial networks and achieve comparable performance with only 20% of the training data on CIFAR-10. Furthermore, when combined with StyleGAN2-ADA, it enhances the Frechet Inception Distance (FID) results on CIFAR-10, LSUN-CAT, and FFHQ-256 datasets.
COMPUTER VISION AND IMAGE UNDERSTANDING
(2023)
Article
Computer Science, Artificial Intelligence
Huiting Liu, Yi Chen, Peipei Li, Peng Zhao, Xindong Wu
Summary: In recent years, review-based methods have been widely used for learning user representations due to the rich information contained in reviews. However, the problem of data sparsity in reviews poses a challenge as few users write reviews. Another approach, using social information based on graph neural networks, is hindered by the unavailability of social graphs in most real-world scenarios. To address these issues, we propose ERUR, a new model that enhances review-based user representations by incorporating learned social graphs. We conduct experiments on seven datasets to demonstrate the effectiveness of ERUR in user representation learning compared to state-of-the-art recommendation models.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Lijia Deng, Shui-Hua Wang, Yu-Dong Zhang
Summary: Cell segmentation and counting are important and time-consuming steps in biomedical research. Traditional counting methods require exact cell locations, but there are few datasets with detailed object coordinates. To overcome this, we propose a GAN-based multi-task model and a novel loss function. Our method achieves excellent results in cell counting and segmentation, and significantly improves image processing speed.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Bin Li, Adib Keikhosravi, Agnes G. Loeffler, Kevin W. Eliceiri
Summary: The article introduces a deep learning-based single image super-resolution technique that can reconstruct high-resolution histology images from low-resolution inputs, reducing cost and increasing data sharing and acquisition speed. Results show the potential application of this technique in clinical pathology diagnosis.
MEDICAL IMAGE ANALYSIS
(2021)
Article
Computer Science, Information Systems
Lianning Cai, Kaitian Cao, Yongpeng Wu, Yuan Zhou
Summary: Spectrum sensing is a key problem in cognitive radio networks. This letter proposes a spectrogram-aware CNN algorithm (S-CNN) that uses the spectrogram of signal samples as input and includes data augmentation based on a deep convolutional generative adversarial network to improve the CNN model's generalization. Simulation results show that the S-CNN algorithm outperforms CNN and LSTM-based methods in terms of detection performance.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Lei Ding, Hao Tang, Yahui Liu, Yilei Shi, Xiao Xiang Zhu, Lorenzo Bruzzone
Summary: This paper proposes an adversarial shape learning network (ASLNet) to improve the accuracy of building segmentation by modeling the shape patterns of buildings. Experimental results show that the proposed ASLNet achieves significant improvements in both pixel-based accuracy and object-based quality measurements.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Yanjun Peng, Jindong Sun, Yande Ren, Dapeng Li, Yanfei Guo
Summary: This paper proposes a histogram-driven generative adversarial network (HisGAN) to estimate CT images paired with MRI. It uses a dynamic scaling factor to improve the learning of different image styles. The method employs adversarial learning, multiple learnable parameters, and deep residual networks to enhance the generation of high-quality medical images.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Engineering, Biomedical
Jiaxin Xie, Siyu Chen, Yongqing Zhang, Dongrui Gao, Tiejun Liu
Summary: An algorithm combining LGANs and MoCNN for MI classification, along with an attention network for model performance improvement, was proposed. Generating more realistic data and expanding the training set improved the classification model’s performance.
JOURNAL OF NEURAL ENGINEERING
(2021)
Article
Engineering, Industrial
Ji-Yan Wu, Min Wu, Zhenghua Chen, Xiaoli Li, Ruqiang Yan
Summary: This study proposes a joint classification-regression scheme for multi-stage RUL prediction, which classifies system health stages based on real-time sensory data and trained models, and performs stage-level RUL prediction with regression algorithms. Experimental results show that the proposed method achieves approximately 6.5% accuracy improvement over state-of-the-art algorithms in RUL prediction.
JOURNAL OF MANUFACTURING SYSTEMS
(2021)
Article
Computer Science, Interdisciplinary Applications
Xiangyu Zhang, Jianqing Li, Zhipeng Cai, Li Zhang, Zhenghua Chen, Chengyu Liu
Summary: In this study, two potential training strategies were explored to address the over-fitting problem in AF detection. The strategies involved using FFT and Hanning-window-based filter to suppress individual differences, as well as training the model on wearable ECG data to improve robustness. These strategies significantly enhanced the detection accuracy rates of the resulting deep networks for AF detection.
MEDICAL & BIOLOGICAL ENGINEERING & COMPUTING
(2021)
Article
Automation & Control Systems
Jie Ding, Changyun Wen, Guoqi Li, Zhenghua Chen
Summary: In this paper, the problem of finding key nodes in a network is approximately solved by proposing three algorithms step by step. By relaxing Boolean constraints, a convex problem is obtained and an inexact alternating direction method of multipliers (IADMMs) is proposed. An extension method called degree-based IADMM (D-IADMM) is introduced based on the degree distribution to pinpoint key nodes, and further performance enhancement is achieved with local optimization in LD-IADMM. The effectiveness of the proposed algorithms is validated on various networks ranging from Erdos-Renyi networks to real-life networks.
IEEE TRANSACTIONS ON CYBERNETICS
(2021)
Article
Automation & Control Systems
Zhenghua Chen, Min Wu, Rui Zhao, Feri Guretno, Ruqiang Yan, Xiaoli Li
Summary: This article proposes an attention-based deep learning framework for the prediction of machine's remaining useful life (RUL). By integrating handcrafted features with automatically learned features and developing a feature fusion framework, the performance of RUL prediction can be improved.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2021)
Article
Environmental Sciences
He Xin, Chen Zhenghua, Lu Yongqiang, Zhang Wei, Yu Kefu
Summary: The study investigated the seasonal and interannual variabilities of sea surface wind in coral reef regions in the South China Sea, revealing both consistencies and differences in wind speed changes among different zones. Regions with high SSW are typically associated with low SST, suggesting a potential inverse relationship between SSW and SST in coral reef areas.
CHINESE GEOGRAPHICAL SCIENCE
(2021)
Article
Computer Science, Artificial Intelligence
Yaping Fu, Yushuang Hou, Zhenghua Chen, Xujin Pu, Kaizhou Gao, Ali Sadollah
Summary: This study proposes an integrated production and distribution optimization problem, utilizing a mixed integer programming model and an enhanced black widow optimization algorithm. The performance and efficiency of the designed method are validated through extensive experiments, showing its superiority over other optimizers. Experimental analysis on sensitive parameters and comparisons with well-known optimizers are conducted to further demonstrate the effectiveness of the proposed method.
SWARM AND EVOLUTIONARY COMPUTATION
(2022)
Article
Computer Science, Information Systems
Chao Tang, Dajian Zhou, Lihua Dou, Chaoyang Jiang
Summary: This study presents a 3D range-only localization algorithm based on improved unscented Kalman filtering. The algorithm can determine the location of unknown UWB nodes in a 3D environment through a moving node with low computational complexity, aiding agents in accurately identifying feature points. The algorithm, validated through theoretical analysis, numerical simulations, and physical experiments, reduces computational burden while maintaining system stability and accuracy.
Article
Engineering, Civil
Xitao Wu, Chao Wei, Hanqing Tian, Weida Wang, Chaoyang Jiang
Summary: This paper presents a fault-tolerant controller for the path-following of independently actuated electric autonomous vehicles with steer-by-wire systems. The controller can handle steering motor faults and disturbances in the vehicle dynamic system. A tube-based model predictive control framework is introduced to ensure vehicle stability and tracking performance. The effectiveness of the controller is demonstrated through joint simulation and real-vehicle experiments.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Xingqi Wang, Chaoyang Jiang, Shuxuan Sheng, Yanjie Xu
Summary: This paper proposes a stop-line-aided cooperative positioning framework for VANET in intersection scenarios, which utilizes V2V communication and stop line information to enhance the positioning performance. A self-positioning correction scheme and extended Kalman filter are used to fuse the local observations and inter-vehicle distance measurements. The framework improves the positioning performance of VANET and experiments in Beijing demonstrate its effectiveness.
IEEE TRANSACTIONS ON INTELLIGENT VEHICLES
(2023)
Article
Automation & Control Systems
Hanqing Tian, Chao Wei, Chaoyang Jiang, Zirui Li, Jibin Hu
Summary: In this article, a new personalized planning and control approach for lane change assistance system is proposed. The approach learns a driver-specific lane-changing policy through end-to-end imitation learning from a few demonstrations. A novel learnable predictive model of the vehicle-driver system is built and an adaptable cost function for the lane change controller is designed. The approach outperforms model-free learning approach in terms of imitation accuracy, interpretability, data efficiency, and generalized performance.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2023)
Article
Mechanics
Alistair D. G. Hales, Lorna J. Ayton, Chaoyang Jiang, Ahmed Mahgoub, Roman Kisler, Rowena Dixon, Charitha de Silva, Danielle Moreau, Con Doolan
Summary: This paper examines the impact of anisotropic turbulence on the generation of noise from the interaction between an aerofoil and turbulence at the leading edge. The study uses thin aerofoil theory to model the aerofoil as a semi-infinite plate and solves the scattering of incoming turbulence using the Wiener-Hopf technique. The authors develop a novel axisymmetric wavenumber-frequency model that captures the wall-normal variation in turbulence characteristics.
JOURNAL OF FLUID MECHANICS
(2023)
Article
Engineering, Electrical & Electronic
Yinqiu Xia, Chengpu Yu, Chaoyang Jiang
Summary: This article studies the problem of distributed localization for a sensor network using noisy distance measurements. A coordinate descent scheme-based method is developed for solving the localization problem, which guarantees convergence and can reach the globally optimal solution. Simulation examples demonstrate the effectiveness of the proposed method in both noise-free and noisy scenarios.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Review
Mathematical & Computational Biology
Kai Cheng, Lixia Li, Yanmin Du, Jiangtao Wang, Zhenghua Chen, Jian Liu, Xiangsheng Zhang, Lin Dong, Yuanyuan Shen, Zhenlin Yang
Summary: Percutaneous puncture is a common medical procedure that can benefit from image guidance and surgical robot-assistance. These technologies have shown potential in improving accuracy and precision of percutaneous procedures, reducing radiation exposure, and lowering complication risks. However, challenges such as integration, perception, and needle insertion deviation need further research to optimize the utilization of these technologies in clinical practice.
MATHEMATICAL BIOSCIENCES AND ENGINEERING
(2023)
Proceedings Paper
Automation & Control Systems
Xiaoni Zheng, Xuetong Ye, Zhe Jin, Tianyan Lan, Chaoyang Jiang
Summary: We propose a dense mapping algorithm SLD-MAP based on surfel with line constraint for room-scale and urban-size environment, aiming to improve reconstruction accuracy and reduce void space on the reconstruction surface. The algorithm utilizes visual odometry to estimate camera poses and reconstructs the 3D environment using input depth and RGB images.
2022 17TH INTERNATIONAL CONFERENCE ON CONTROL, AUTOMATION, ROBOTICS AND VISION (ICARCV)
(2022)
Proceedings Paper
Engineering, Industrial
Keyu Wu, Zhenghua Chen, Min Wu, Shili Xiang, Ruibing Jin, Le Zhang, Xiaoli Li
Summary: This paper proposes a method to improve the generalization capability of reinforcement learning algorithms through multi-task self-supervised adaptation. The method extracts high-level feature representations by incorporating multiple self-supervision tasks and achieves state-of-the-art results on complex tasks without increasing inference time.
2022 IEEE 17TH CONFERENCE ON INDUSTRIAL ELECTRONICS AND APPLICATIONS (ICIEA)
(2022)
Article
Construction & Building Technology
Samiran Khorat, Debashish Das, Rupali Khatun, Sk Mohammad Aziz, Prashant Anand, Ansar Khan, Mattheos Santamouris, Dev Niyogi
Summary: Cool roofs can effectively mitigate heatwave-induced excess heat and enhance thermal comfort in urban areas. Implementing cool roofs can significantly improve urban meteorology and thermal comfort, reducing energy flux and heat stress.
ENERGY AND BUILDINGS
(2024)
Article
Construction & Building Technology
Qi Li, Jiayu Chen, Xiaowei Luo
Summary: This study focuses on the vertical wind conditions as a main external factor that limits the energy assessment of high-rise buildings in urban areas. Traditional tools for energy assessment of buildings use a universal vertical wind profile estimation, without taking into account the unique wind speed in each direction induced by the various shapes and configurations of buildings in cities. To address this limitation, the study developed an omnidirectional urban vertical wind speed estimation method using direction-dependent building morphologies and machine learning algorithms.
ENERGY AND BUILDINGS
(2024)
Article
Construction & Building Technology
Xiaojun Luo, Lamine Mahdjoubi
Summary: This paper presents an integrated blockchain and machine learning-based energy management framework for multiple forms of energy allocation and transmission among multiple domestic buildings. Machine learning is used to predict energy generation and consumption patterns, and the proposed framework establishes optimal and automated energy allocation through peer-to-peer energy transactions. The approach contributes to the reduction of greenhouse gas emissions and enhances environmental sustainability.
ENERGY AND BUILDINGS
(2024)
Article
Construction & Building Technology
Ying Yu, Yuanwei Xiao, Jinshuai Chou, Xingyu Wang, Liu Yang
Summary: This study proposes a dual-layer optimization design method to maximize the energy sharing potential, enhance collaborative benefits, and reduce the storage capacity of building clusters. Case studies show that the proposed design significantly improves the performance of building clusters, reduces energy storage capacity, and shortens the payback period.
ENERGY AND BUILDINGS
(2024)
Article
Construction & Building Technology
Felix Langner, Weimin Wang, Moritz Frahm, Veit Hagenmeyer
Summary: This paper compares two main approaches to consider uncertainties in model predictive control (MPC) for buildings: robust and stochastic MPC. The results show that compared to a deterministic MPC, the robust MPC increases the electricity cost while providing complete temperature constraint satisfaction, while the stochastic MPC slightly increases the electricity cost but fulfills the thermal comfort requirements.
ENERGY AND BUILDINGS
(2024)
Article
Construction & Building Technology
Somil Yadav, Caroline Hachem-Vermette
Summary: This study proposes a mathematical model to evaluate the performance of a Double Skin Facade (DSF) system and its impact on indoor conditions. The model considers various design parameters and analyzes their effects on the system's electrical output and room temperature.
ENERGY AND BUILDINGS
(2024)
Article
Construction & Building Technology
Ruijun Chen, Holly Samuelson, Yukai Zou, Xianghan Zheng, Yifan Cao
Summary: This research introduces an innovative resilient design framework that optimizes building performance by considering a holistic life cycle perspective and accounting for climate projection uncertainties. The study finds that future climate scenarios significantly impact building life cycle performance, with wall U-value, windows U-value, and wall density being major factors. By using ensemble learning and optimization algorithms, predictions for carbon emissions, cost, and indoor discomfort hours can be made, and the best resilient design scheme can be selected. Applying this framework leads to significant improvements in building life cycle performance.
ENERGY AND BUILDINGS
(2024)