Article
Neurosciences
Sina L. Mansour, Ye Tian, B. T. Thomas Yeo, Vanessa Cropley, Andrew Zalesky
Summary: This study explores high-resolution connectomes independent of brain parcellation atlas, proposing new methodologies that are computationally feasible and improving neural fingerprinting and behavior prediction. The findings reveal that individual uniqueness in cortical gradients is located in association cortices, with functional connectivity being a more accurate predictor of behavior compared to differentiators of identity, such as cortical curvature.
Article
Computer Science, Hardware & Architecture
Zexin Li, Yuqun Zhang, Ao Ding, Husheng Zhou, Cong Liu
Summary: The paper introduces a theoretical framework and practical mapping algorithms for solving computation and data mapping problems in GPU-based embedded systems. Experimental results show that these algorithms can achieve faster completion times compared to state-of-the-art techniques, and perform consistently well across different workloads.
JOURNAL OF SYSTEMS ARCHITECTURE
(2021)
Article
Engineering, Electrical & Electronic
Ning Chu, Han Zhao, Liang Yu, Qian Huang, Yue Ning
Summary: This study proposes a novel method based on convolution approximation and GPU platform to quickly obtain high-resolution acoustic imaging. The results show that the separable invariant convolution kernel provides the fastest imaging speed, while the variant kernel has the highest time consumption.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2021)
Article
Computer Science, Information Systems
Chanyoung Oh, Saehanseul Yi, Jongkyu Seok, Hyeonjin Jung, Illo Yoon, Youngmin Yi
Summary: This paper proposes a CPU-GPU hybrid scheduling algorithm in Hadoop that fully utilizes the CPU and GPU in a node in an adaptive manner, with the help of a GPU monitor to adjust the Container number. Experimental results show that the proposed scheduling algorithm achieves an average speedup of 3.87x compared to a CPU-only Hadoop.
CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Tobias Groth, Sven Groppe, Thilo Pionteck, Franz Valdiek, Martin Koppehel
Summary: Modern computer systems can achieve significant performance improvements by using various types of hardware acceleration. Different accelerators require different optimized data structures and memory configurations to achieve the best performance. APUs, which combine CPU and integrated GPU, support shared memory and enable CPU and iGPU collaboration on pointer-based data structures.
KNOWLEDGE AND INFORMATION SYSTEMS
(2023)
Article
Computer Science, Interdisciplinary Applications
Varsha Sreenivasan, Sawan Kumar, Franco Pestilli, Partha Talukdar, Devarajan Sridharan
Summary: Different tractography algorithms can produce widely varying connectivity estimates. This study introduces a GPU-based implementation that achieves significantly faster speeds and improved accuracy in generating brain connectomes. The implementation also has potential applications in various real-world problems.
NATURE COMPUTATIONAL SCIENCE
(2022)
Article
Neurosciences
Sanjay Ghosh, Ashish Raj, Srikantan S. Nagarajan
Summary: This article introduces a computational framework that reconstructs functional connectivity from structural connectivity by identifying a joint subspace of eigenmodes. It is found that a small number of these eigenmodes are sufficient for reconstruction and the proposed algorithm shows competitive performance and better interpretability compared to existing methods.
Article
Computer Science, Hardware & Architecture
Binbin Zhou, Lu Lu
Summary: This paper introduces an efficient 3D FFT framework for multi-GPU distributed-memory systems, which utilizes a hybrid programming model combining MPI and OpenMP for effective communication, and adopts an asynchronous strategy and fast parallel kernels for acceleration.
JOURNAL OF SUPERCOMPUTING
(2022)
Article
Computer Science, Information Systems
Jian Feng, Kaihong Song, Ming Fang, Wei Chen, Guoda Xie, Zhixiang Huang, Xianliang Wu
Summary: This paper proposes a numerical method based on finite-difference time-domain scheme to simulate the plasma formation process in high-power microwave devices, and efficiently accelerates the algorithm using heterogeneous computing technique. It provides an efficient method for investigating the phenomenon of high-power microwave air breakdown.
Article
Environmental Sciences
Ping Zhang, Yongchao Zhang, Deqing Mao, Jianan Yan, Shuaidi Liu
Summary: This paper introduces an improved Poisson distribution-based maximum likelihood (IPML) method that effectively improves the algorithm convergence speed without high-dimensional matrix operations. A GPU-based parallel processing architecture is proposed, along with a cooperative CPU-GPU working model, achieving parallel optimization of echo reception, preprocessing, and super-resolution processing. The proposed method significantly improves computational efficiency without sacrificing performance, as verified using real dataset.
Article
Computer Science, Information Systems
Fei Yin, Feng Shi
Summary: This paper integrates the Kriging proxy model and energy efficiency modeling method into a cluster optimization algorithm of CPU and GPU hybrid architecture, proposing a parallel computing model for large-scale CPU/GPU heterogeneous high-performance computing systems and providing algorithm optimization for CPU/GPU heterogeneous clusters. The experimental results show significant speedup ratios for constructing the Kriging proxy model and the search algorithm, demonstrating the high feasibility of this heterogeneous cluster optimization algorithm.
CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Trung-Hieu Tran, Kaicong Sun, Sven Simon
Summary: This paper proposes a GPU-accelerated computational framework for reconstructing high-resolution light-field images under mixed noise conditions. The framework combines a joint data fidelity term and weighted non-local total variation approach for regularization. Experimental results demonstrate that the proposed approach achieves better reconstruction quality and overcomes the limitations of previous work in handling large-scale tasks.
JOURNAL OF REAL-TIME IMAGE PROCESSING
(2022)
Article
Computer Science, Interdisciplinary Applications
Andreas Buttinger-Kreuzhuber, Artem Konev, Zsolt Horvath, Daniel Cornel, Ingo Schwerdorf, Guenter Bloeschl, Juergen Waser
Summary: This paper presents an integrated modeling framework for accurate predictions of flood hazard from heavy rainfalls. By integrating complementary models and utilizing GPU acceleration, the accuracy and simulation time of the model are improved. The framework is validated and tested in various scenarios, showing significant acceleration and the ability to simulate a large urban area in real-time.
ENVIRONMENTAL MODELLING & SOFTWARE
(2022)
Article
Agriculture, Multidisciplinary
Ulrich Johan Isachsen, Theoharis Theoharis, Ekrem Misimi
Summary: The study introduces an efficient 3D registration algorithm based on ICP for eye-in-hand configuration using an RGB-D camera, with faster GPU implementation and high resolution. Results indicate that the point-to-plane linear least squares optimizer provides the best performance in terms of accuracy and efficiency.
COMPUTERS AND ELECTRONICS IN AGRICULTURE
(2021)
Article
Computer Science, Information Systems
Xinjian Long, Xiangyang Gong, Bo Zhang, Huiyang Zhou
Summary: This paper proposes a novel framework for UVM oversubscription management in discrete CPU-GPU systems, which significantly outperforms the state-of-the-art methods on memory-intensive benchmarks. It reduces page thrashing and achieves improved IPC under different levels of memory oversubscription.
JOURNAL OF GRID COMPUTING
(2023)
Article
Computer Science, Hardware & Architecture
Kai Zhong, Xuefei Ning, Guohao Dai, Zhenhua Zhu, Tianchen Zhao, Shulin Zeng, Yu Wang, Huazhong Yang
Summary: The study proposes a low-bit training framework to improve the efficiency of training CNNs. By adopting techniques like element-wise scaling and group-wise scaling, the challenges of using low-bit integer format in training have been successfully addressed. Experimental results show that the framework achieves a superior tradeoff between accuracy and bit-width.
IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
(2022)
Editorial Material
Neurosciences
Mingrui Xia, Yong He
BIOLOGICAL PSYCHIATRY
(2023)
Article
Neurosciences
Zilong Zeng, Tengda Zhao, Lianglong Sun, Yihe Zhang, Mingrui Xia, Xuhong Liao, Jiaying Zhang, Dinggang Shen, Li Wang, Yong He
Summary: In this study, a new 3D mixed-scale asymmetric convolutional segmentation network (3D-MASNet) was proposed for tissue segmentation of 6-month-old infant brain MRI images. Compared to traditional single-scale symmetric convolutions, this approach demonstrated better accuracy and achieved the best performance in the evaluation.
HUMAN BRAIN MAPPING
(2023)
Article
Automation & Control Systems
Jing Zhang, Yuan Shen, Yu Wang, Xudong Zhang, Jian Wang
Summary: Edge computing is important for future Internet of Things systems, as it can reduce service latency and energy consumption by offloading computational tasks to edge servers. Caching appropriate services in the edge server can improve the quality of service, but it requires joint optimization of resource allocation considering different timescales of caching and offloading operations. This article proposes a novel hierarchical deep reinforcement learning scheme to optimize collaborative service caching and computation offloading.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Neurosciences
Yuting Li, Xiaofei Huang, Xiuhang Ruan, Dingna Duan, Yihe Zhang, Shaode Yu, Amei Chen, Zhaoxiu Wang, Yujian Zou, Mingrui Xia, Xinhua Wei
Summary: This study aimed to predict the occurrence of freezing of gait (FOG) in early drug-naive Parkinson's disease (PD) patients using machine learning, and investigate alterations in cerebral morphology in early PD. The study found that models trained with structural morphological features showed good predictive performance for FOG, and adding clinical and laboratory data improved the performance. Elastic net-support vector machine models performed the best, and the main features used for prediction were structural morphological features mainly distributed in the left cerebrum. The study also found that the bilateral olfactory cortex showed significant cortical expansion in PD patients.
NPJ PARKINSONS DISEASE
(2022)
Article
Engineering, Electrical & Electronic
Wenjun Tang, Mingyen Lee, Juejian Wu, Yixin Xu, Yao Yu, Yongpan Liu, Kai Ni, Yu Wang, Huazhong Yang, Vijaykrishnan Narayanan, Xueqing Li
Summary: Bitwise logic-in-memory (BLiM) is a promising approach to efficient computing in data-intensive applications. This work proposes a new BLiM approach based on ferroelectric field-effect transistors (FeFETs), supporting various computing functions and achieving higher energy efficiency and speed. The evaluation shows significant improvements in latency and energy consumption for typical operations, such as in-memory XOR and the advanced encryption standard (AES).
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2023)
Article
Computer Science, Hardware & Architecture
Jingbo Hu, Guohao Dai, Liuzheng Wang, Liyang Lai, Yu Huang, Huazhong Yang, Yu Wang
Summary: This paper proposes an adaptive multidimensional parallel fault simulation framework based on the CPU-GPU heterogeneous system. It addresses the challenges of path divergence, unbalanced workload, and poor scalability, and further accelerates by introducing a 4-D parallel architecture on multiple GPUs. Experimental results show that compared to the commercial tool, the fault simulator based on 8 GPUs achieves an average speedup of 105.7 times, and for millions of gate-level circuits, the fault simulator based on one GPU achieves a speedup of up to 25.9 times compared to the CPU single-threaded simulator.
IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
(2023)
Article
Computer Science, Hardware & Architecture
Shulin Zeng, Guohao Dai, Niansong Zhang, Xinhao Yang, Haoyu Zhang, Zhenhua Zhu, Huazhong Yang, Yu Wang
Summary: This paper proposes the H3M framework to optimize the architecture, scheduling, and mapping for INFaaS on cloud FPGA. H3M outperforms other accelerators in terms of EDP reduction on the ASIC platform. On the Xilinx U200 and U280 FPGA platforms, H3M significantly reduces EDP compared to Herald.
IEEE TRANSACTIONS ON COMPUTERS
(2023)
Article
Clinical Neurology
Lei Wang, Qing Ma, Xiaoyi Sun, Zhilei Xu, Jiaying Zhang, Xuhong Liao, Xiaoqin Wang, Dongtao Wei, Yuan Chen, Bangshan Liu, Chu-Chung Huang, Yanting Zheng, Yankun Wu, Taolin Chen, Yuqi Cheng, Xiufeng Xu, Qiyong Gong, Tianmei Si, Shijun Qiu, Ching-Po Lin, Jingliang Cheng, Yanqing Tang, Fei Wang, Jiang Qiu, Peng Xie, Lingjiang Li, Yong He, Mingrui Xia, Yihe Zhang
Summary: This study conducted frequency-resolved connectome analysis on a large sample of MDD patients and healthy controls, revealing significant frequency-dependent connectome alterations in MDD. These alterations mainly occur in the left parietal, temporal, precentral, and fusiform cortices, as well as bilateral precuneus. Additionally, the connectome alteration in the high frequency band (0.16-0.24 Hz) is significantly associated with illness duration.
JOURNAL OF AFFECTIVE DISORDERS
(2023)
Article
Neurosciences
Junling Wang, Lianglong Sun, Lili Chen, Junyan Sun, Yapei Xie, Dezheng Tian, Linlin Gao, Dongling Zhang, Mingrui Xia, Tao Wu
Summary: Neuroimaging studies have shown that dysfunction of the amygdala plays a crucial role in the non-motor symptoms of Parkinson's disease. However, the specific relationship between amygdala subregions and these symptoms has not been well-defined. Using resting-state functional MRI, researchers found that the amygdala subregions in Parkinson's disease exhibited altered functional connectivity, particularly with the frontal, temporal, insular cortex, and putamen. Each subregion also displayed distinct hypoconnectivity with different limbic systems, and this hypoconnectivity was associated with various non-motor symptoms such as emotion, pain, olfaction, cognition, and sleepiness. These findings provide new insights into the pathogenesis of non-motor symptoms in Parkinson's disease.
NPJ PARKINSONS DISEASE
(2023)
Correction
Neurosciences
Junling Wang, Lianglong Sun, Lili Chen, Junyan Sun, Yapei Xie, Dezheng Tian, Linlin Gao, Dongling Zhang, Mingrui Xia, Tao Wu
NPJ PARKINSONS DISEASE
(2023)
Article
Neurosciences
Yuxing Hao, Huashuai Xu, Mingrui Xia, Chenwei Yan, Yunge Zhang, Dongyue Zhou, Tommi Karkkainen, Lisa D. Nickerson, Huanjie Li, Fengyu Cong
Summary: This study proposes an effective and powerful harmonisation strategy based on dual-projection (DP) theory of independent component analysis (ICA) to remove scanner/site effects while preserving signals of interest. The method shows superior performance compared to GLM-based and conventional ICA harmonisation methods in both simulation and real datasets.
EUROPEAN JOURNAL OF NEUROSCIENCE
(2023)
Article
Computer Science, Hardware & Architecture
Hanbo Sun, Zhenhua Zhu, Chenyu Wang, Xuefei Ning, Guohao Dai, Huazhong Yang, Yu Wang
Summary: This paper introduces an efficient co-exploration framework, named Gibbon, for NN models and PIM architectures. It improves search efficiency through a carefully designed co-exploration space and an evolutionary search algorithm, ESAPP, and addresses the issue of time-consuming evaluation with a multilevel joint simulator. Experimental results show that Gibbon can find better NN models and PIM architectures in a short amount of time, improving the accuracy of co-design results and reducing the energy-delay-product.
IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Xuefei Ning, Yin Zheng, Zixuan Zhou, Tianchen Zhao, Huazhong Yang, Yu Wang
Summary: Neural architecture search (NAS) can automatically discover well-performing architectures in a large search space and has been shown to bring improvements to various applications. To improve the sample efficiency of search space exploration, GATES++ incorporates multifaceted information about NN's operation-level and architecture-level computing semantics into its construction and training, and it can discover better architectures after evaluating the same number of architectures.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Robotics
Zihan Lin, Jincheng Yu, Lipu Zhou, Xudong Zhang, Jian Wang, Yu Wang
Summary: This paper proposes a novel framework using low-cost sensors and algorithms to detect changes in a point cloud map, and creates a corresponding dataset and metrics for evaluation. Experiments show that the framework can effectively detect changes in the dataset.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)