A5101891026- Optimization and Variational Analysis
- Parallel Computing and Optimization Techniques
- Nonlinear Partial Differential Equations
- Advanced Data Storage Technologies
- Nonlinear Differential Equations Analysis
- Distributed and Parallel Computing Systems
- Face and Expression Recognition
- Cooperative Communication and Network Coding
- Interconnection Networks and Systems
- Neural Networks and Applications
- Advanced Optimization Algorithms Research
- Image and Video Quality Assessment
- Speech and Audio Processing
- Advanced Database Systems and Queries
- Fixed Point Theorems Analysis
- Stability and Controllability of Differential Equations
- Advanced Mathematical Modeling in Engineering
- Geometric Analysis and Curvature Flows
- Embedded Systems Design Techniques
- Advanced Image Processing Techniques
- Caching and Content Delivery
- Service-Oriented Architecture and Web Services
- Advanced Banach Space Theory
- Advanced Image Fusion Techniques
- Semantic Web and Ontologies
Taizhou University
2008-2025
Sichuan University
2013-2024
Ghent University
2023
Ghent University Hospital
2023
Quanzhou Normal University
2022
National University of Defense Technology
2017-2019
Beijing University of Posts and Telecommunications
2016-2018
Southwestern University of Finance and Economics
2014
National Natural Science Foundation of China
2014
Xingyi Normal University for Nationalities
2011
This article introduces the photonic network-on-wafer graphics processing unit (GPU) architecture to overcome fundamental limitations in electrical interconnect scaling by implementing inter-GPU network a wafer-scale optical interposer. We argue that photonic-NoW GPU is scalable architecture, delivering significant performance benefits power-efficient manner.
Bandwidth non-uniformity in multi-chip GPUs poses a major design challenge for its last-level cache (LLC) architecture. Whereas memory-side LLC caches data from the local memory partition while being accessible by all chips, an SM-side is private to chip caching partitions. We find that some workloads prefer others LLC, and this preference solely depends on which organization maximizes effective bandwidth. In contrast prior work optimizes bandwidth beyond we make observation ahead of...
This paper introduces a coin recognition method with rotation invariance. The invariance feature is represented by the absolute value of Fourier coefficients polar image on circles different radii. Moreover, approximation used to reduce variations surface such as light reflection effect. Then coins can be distinguished feeding those features into multi-layered BP neural network. Finally experiments are given show effectiveness proposed method.
In this note, we try to generalize the classical Cauchy-Lipschitz-Picard theorem on global existence and uniqueness for Cauchy initial value problem of ordinary differential equation with Lipschitz condition, weaken condition. We can also get uniqueness.
Fuzzy information granulation transfers the time series analysis from numerical platform to granular platform, which enables us study at a different granularity. In previous studies, each fuzzy granule in can reflect average, range, and linear trend characteristics of data corresponding window. order get more general granule, this paper proposes polynomial granules, both nonlinear The distance metric proposed granules is given theoretically. After studying measure its geometric...
MCM-GPUs scale performance by integrating multiple chiplets within the same package. How to partition aggregate compute resources across poses a fundamental trade-off in versus cost and sustainability. We propose <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Performance Per Wafer (PPW)</i> metric explore this we find that while is maximized with few large chiplets, environmental footprint minimized many small optimum balance achieved...
UE-to-Network relay leverages the proximity communication between user equipments (UEs) and allows certain UEs to provide assistance for others, which can greatly improve system energy efficiency. In this paper, we consider scenario where suffering from bad channel condition low battery level communicate with base station directly or via help of other in heterogeneous networks. The optimal power allocation connectivity among are studied, aims at minimizing transmission while guaranteeing...
为了有效提高语音情感识别的性能，需要对嵌入在高维声学特征空间的非线性流形上的语音特征数据作非线性降维处理。监督局部线性嵌入(SLLE)是一种典型的用于非线性降维的监督流形学习算法。该文针对SLLE存在的缺陷，提出一种能够增强低维嵌入数据的判别力，具备最优泛化能力的改进SLLE算法。利用该算法对包含韵律和音质特征的48维语音情感特征数据进行非线性降维，提取低维嵌入判别特征用于生气、高兴、悲伤和中性4类情感的识别。在自然情感语音数据库的实验结果表明，该算法仅利用较少的9维嵌入特征就取得了90.78%的最高正确识别率，比SLLE提高了15.65%。可见，该算法用于语音情感特征数据的非线性降维，可以较好地改善语音情感识别结果。
Despite the increasing investment in integrated GPUs and next-generation interconnect research, discrete connected by PCI Express still account for dominant position of market, management data communication between CPU GPU continues to evolve. Initially, programmer controls transfer explicitly. To simplify programming enable system-wide atomic memory operations, vendors have developed a model that provides single virtual address space. The page migration engine this migrates pages on demand...
A new method of speech emotion recognition in signal via Fuzzy Least Squares Support Vector Machines (FLSSVM) is proposed for recognition. Based on extracting prosody and voice quality features from emotional speech, FLSSVM used to construct the optimum separating hyperplane realize recognizing four main Chinese including anger, happiness, sadness surprise. Compared with other present methods recognition, computer simulation results show that can achieve higher average correct rate better...
In this work the minimization problem for difference of convex (DC) functions is studied by using Moreau envelopes and descent method with gradient employed to approximate numerical solution. The main regularization idea in inspired Hiriart-Urruty [14], Moudafi[17], regularize components DC adapting different parameters strategic matrices flexibly evaluate whole problem. It shown that inertial as well classic scheme tend towards an approximation stationary point original
The demand for greater computing power has driven the development of Multi-chip-module GPUs (MCM-GPUs), which greatly improve parallel processing capabilities. Unfortunately, MCM-GPUs have encountered a notable challenge, performance bottleneck caused by remote accesses through inter-module network. In this work, we found significant data access redundancy among SMs within GPU module can be coalesced to reduce network pressure. However, how design coalescing scheme identify memory addresses...
<title>Abstract</title> The tone-mapping operator (TMO) aims to convert high dynamic range images into low images, enabling them be displayed on standard monitors. However, this conversion process inevitably results in a decrease image quality, such as structural damage, contrast degradation and color artifacts, which impacts human’s visual perception. Evaluating the quality of tone-mapped (TMIs) effectively remains challenge. To address issue, paper proposes novel blind metric for...
Caching popular contents at mobile devices can improve the system performance and relieve traffic burden for device-to-device (D2D) enabled ultra dense network (UDN). In this paper, D2D networks are formulated as a 3-layer architecture involving link layer, social tie layer preference layer. Based on architecture, proactive content sharing scheme including caching, diffusion resource allocation is proposed. For we propose joint seed user (DUE) caching selection algorithm (JSCSA) with graph...
At present, graphics processing units (GPUs) has been widely used for scientific and high performance acceleration in the general purpose computing area, which is inseparable from SIMT (Single-Instruction, Multiple-Thread) execution model. With SIMT, GPUs can fully utilize advantages of SIMD parallel computing. However, when threads a warp do not follow same path, control divergence generates affects hardware utilization. In response to this problem, regrouping method proposed combine...