检索结果-南通市图书馆

Re-quantization based binary graph neural networks

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第7期67卷 160-171页

作者： Kai-Lang YAO wu-jun li National Key Laboratory for Novel Software Technology Department of Computer Science and TechnologyNanjing University

Binary neural networks have become a promising research topic due to their advantages of fast inference speed and low energy consumption. However, most existing studies focus on binary convolutional neural networks, while less attention has been paid to binary graph neural networks. A common drawback of existing studies on binary graph neural networks is that they still include lots of inefficient full-precision operations in multiplying three matrices and are therefore not efficient enough. In this paper, we propose a novel method, called re-quantization-based binary graph neural networks(RQBGN), for binarizing graph neural networks. Specifically, re-quantization, a necessary procedure contributing to the further reduction of superfluous inefficient full-precision operations, quantizes the results of multiplication between any two matrices during the process of multiplying three matrices. To address the challenges introduced by requantization, in RQBGN we first study the impact of different computation orders to find an effective one and then introduce a mixture of experts to increase the model capacity. Experiments on five benchmark datasets show that performing re-quantization in different computation orders significantly impacts the performance of binary graph neural network models, and RQBGN can outperform other baselines to achieve state-of-the-art performance.

关键词： graph neural networks binary neural networks mixture of experts computation-efficient algorithms

Investigation of Endwall Flow Behavior with Plasma Flow Control on a Highly Loaded Compressor Cascade

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Journal of Thermal Science 2012年第4期21卷 295-301页

作者： Xiaohu ZHAO Yinghong li Yun wu jun li Science and Technology on Plasma Dynamics Laboratory Air Force Engineering University Xi' an 710038 China

To discover the flow behavior in the endwall region and mechanism of plasma flow control on a highly loaded compressor cascade, distributions of static pressure coefficient, total pressure loss coefficient and streamline pat- tern were investigated. Results show that cross flow from the pressure surface to neighboring suction surface ex- ists under pitch-wise pressure gradient. The deflected endwall boundary layer flow interacts with the incoming flow, and then both of them leave off the endwall in tile form of a span-wise vortex. Effect of angle of attack on static pressure is greater than that of free stream velocity. The distinct variations of total pressure loss with end- wall actuations are mainly located within the outer verge of a triangular area with high total pressure loss. Effect of pitch-vAse actuation on separated flows is much better than that of stream-wise actuation, and both enhance with the increase of angle of attack and actuation strength. An efficient method for plasma flow control in the endwall region is the increase of actuation strength, such as adjusting discharge voltage or changing plasma power supply.

关键词： plasma flow control compressor cascade endwall region vortex

On the convergence and improvement of stochastic normalized gradient descent

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2021年第3期64卷 105-117页

作者： Shen-Yi ZHAO Yin-Peng XIE wu-jun li National Key Laboratory for Novel Software Technology Department of Computer Science and TechnologyNanjing University

Non-convex models, like deep neural networks, have been widely used in machine learning applications. Training non-convex models is a difficult task owing to the saddle points of models. Recently,stochastic normalized gradient descent(SNGD), which updates the model parameter by a normalized gradient in each iteration, has attracted much attention. Existing results show that SNGD can achieve better performance on escaping saddle points than classical training methods like stochastic gradient descent(SGD).However, none of the existing studies has provided theoretical proof about the convergence of SNGD for non-convex problems. In this paper, we firstly prove the convergence of SNGD for non-convex ***, we prove that SNGD can achieve the same computation complexity as SGD. In addition, based on our convergence proof of SNGD, we find that SNGD needs to adopt a small constant learning rate for convergence guarantee. This makes SNGD do not perform well on training large non-convex models in practice. Hence, we propose a new method, called stagewise SNGD(S-SNGD), to improve the performance of SNGD. Different from SNGD in which a small constant learning rate is necessary for convergence guarantee,S-SNGD can adopt a large initial learning rate and reduce the learning rate by stage. The convergence of S-SNGD can also be theoretically proved for non-convex problems. Empirical results on deep neural networks show that S-SNGD achieves better performance than SNGD in terms of both training loss and test accuracy.

关键词： non-convex problems stochastic normalized gradient descent computation complexity

空位缺陷硅烯纳米带结构与自旋极化电子性质的密度泛函紧束缚研究

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

原子与分子物理学报 2025年第3期42卷 61-66页

作者：扶锦云吴丽君王爽迟中钰刘亚沈阳理工大学理学院沈阳110159

随着纳米电子器件和自旋电子学的不断发展,对小尺寸的纳米硅材料物理和化学性质的研究是研发新型微电子材料器件的基础.硅烯纳米带在实验制备和理论研究中均备受关注.本研究利用基于自洽电荷的密度泛函理论紧束缚方法,研究了不同宽度的... 详细信息

随着纳米电子器件和自旋电子学的不断发展,对小尺寸的纳米硅材料物理和化学性质的研究是研发新型微电子材料器件的基础.硅烯纳米带在实验制备和理论研究中均备受关注.本研究利用基于自洽电荷的密度泛函理论紧束缚方法,研究了不同宽度的硅烯纳米带的结构、稳定性和自旋极化电子性质.研究发现,稳定的完美和带有空位缺陷的硅烯纳米带表现为带有曲翘褶皱的蜂窝结构.纳米带边缘的曲翘度大于中心位置.空位缺陷更容易出现在纳米带边缘.纳米带的宽度可实现完美的硅烯纳米带的半导体性质和金属性之间的转换.硅烯纳米带中出现空位缺陷可有效调控纳米带自旋电子性质从半导体向金属性转变、半导体向半金属性转变,及金属性向为半导体或者半金属性质的转变.这种自旋电子性质的调控手段使得该种纳米材料在纳米电子自旋器件上有着潜在的应用价值.

关键词：硅烯纳米带自旋极化电子性质密度泛函紧束缚空位缺陷

维普期刊数据库博看期刊

Comparison and Optimization of Mid-low Temperature Cogeneration Systems for Flue Gas in Iron and Steel Plants

在线全文

学校读者我要写书评

暂无评论

Journal of Iron and Steel Research International 2013年第11期20卷 33-40页

作者： ZHANG li-hua wu li-jun ZHANG Xiao-hong JU Gui-dong Thermal and Environmental Engineering Institute Tongji University Resources and Environment Department Shandong Water Polytechnic

Three generation systems, namely, steam Rankine cycle （SRC）, organic Rankine cycle （ORC）, and steam-organic combined Rankine cycle （S-ORC）, were simulated using the Engineering Equation Solver fEES） to efficiently utilize flue gas emissions from 200 to 450 ℃ in iron and steel plants. Based on the simulation results for thermal efficiency, exergy efficiency, and power generation, the performances of the three power generation systems were compared and analyzed. To further utilize waste heat from the turbine exhaust steam of the ORC system, cas- cade （）RC （CORC） was designed for heat sources above 300 ℃. Based on a comprehensive performance comparison, the application of the ORC using R141b is preferable for 200 to 300 ℃ flue gas. For 300 to 450 ℃ flue gas, CORC is an alternative technology to improve the efficiency and quality of waste heat utilization. For flue gas above 450 ℃, S-ORC can achieve higher efficiency and power generation than conventional SRC, with a relatively small negative pressure and high dryness of the turbine outlet steam. Hence, S-ORC can be considered as a substitute for SRC.

关键词： mid-low temperature flue gas Rankine cycle thermal energy recovery generation system

Stochastic normalized gradient descent with momentum for large-batch training

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第11期67卷 77-91页

作者： Shen-Yi ZHAO Chang-Wei SHI Yin-Peng XIE wu-jun li National Key Laboratory for Novel Software Technology Department of Computer Science and TechnologyNanjing University

Stochastic gradient descent(SGD) and its variants have been the dominating optimization methods in machine learning. Compared with SGD with small-batch training, SGD with large-batch training can better utilize the computational power of current multi-core systems such as graphics processing units(GPUs)and can reduce the number of communication rounds in distributed training settings. Thus, SGD with large-batch training has attracted considerable attention. However, existing empirical results showed that large-batch training typically leads to a drop in generalization accuracy. Hence, how to guarantee the generalization ability in large-batch training becomes a challenging task. In this paper, we propose a simple yet effective method, called stochastic normalized gradient descent with momentum(SNGM), for large-batch training. We prove that with the same number of gradient computations, SNGM can adopt a larger batch size than momentum SGD(MSGD), which is one of the most widely used variants of SGD, to converge to an?-stationary point. Empirical results on deep learning verify that when adopting the same large batch size,SNGM can achieve better test accuracy than MSGD and other state-of-the-art large-batch training methods.

关键词： non-convex problems large-batch training stochastic normalized gradient descent momentum

Preparation and Characterization of Microcapsules for Self-healing Materials

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Chemical Research in Chinese Universities 2010年第3期26卷 496-500页

作者： liAO Le-ping ZHANG Wei ZHAO Yang li wu-jun National Key Laboratory for Remanufacturing Academy of Armored Force Engineering Beijing 100072 P. R. China

The urea-formaldehyde（UF） capsules filled with a healing agent, a mixture of epoxy resins（epoxy 711 and E-51） as core material, were produced by means of one step in-situ polymerization. The characteristics of these microcapsules were studied via scanning electron microscopy（SEM）, particle size analysis, FTIR and DSC/TGA. The results show that the dispersed and integrated microcapsules of 5 μm in shell thickness were synthesized successfully. The capsules were produced with diameters ranging from 10 to 250 μm via controlling agitation rate. Young＇s modulus of the capsule was a little lower than those of the epoxy resins, but the microcapsules having such a shell thickness were robust enough to survive handling during manufacturing self-healing composites. When damage occurred in the epoxy matrix, the crack could rupture the microcapsule to make the repairing agent release.

关键词： Urea-formaldehyde Microcapsule Epoxy resin Self-healing

Beam steering in planar photonic crystal based on its anomalous dispersive properties

在线全文

学校读者我要写书评

暂无评论

Journal of Zhejiang University-Science A(Applied Physics & Engineering) 2006年第1期7卷 45-54页

作者： wu li-jun MAZILU M GALLETJ．F． KRAUSST．F． The Ultrafast Photonics Collaboration School of Physics ＆ Astronomy University of St Andrews St. Andrews. Fife KY16 9SS. Scotland UK Department of Physics. Hong Kong University of Science and Technology. Hong Kong China

We utilize the anomalous dispersion of planar photonic crystals near the dielectric band edge to control the wavelength-dependent propagation of light. We typically observe an angular swing of up to 10°as the input wavelength is changed from 1290 nm to 1310 rim, which signifies an angular dispersion of 0.5°/am （＂Superprism＂ phenomenon）. Such a strong angular dispersion is of the order required for WDM systems. By tuning the incident angle, light beams with up to 20° divergence were collimated over a 25 nm （1285 nm to 1310 nm） bandwidth using a triangular lattice （＂Supercollimator＂ phenomenon）. The wavelength collimating range can be extended from 25 nm to 40 nm by changing the lattice from triangular to square. These two devices can be realized in the same configuration, simply by tuning the wavelength. Sources of loss are discussed.

关键词： Planar photonic crystals Superprism Supercollimator Anomalous dispersive properties

Effect of Ionic liquid on Detection of Carbofuran by Plant-lipases Inactivation

在线全文

学校读者我要写书评

暂无评论

Chinese Chemical Letters 2006年第11期17卷 1493-1494页

作者： Yan wu li jun HE Wen Jie ZHAO Kui LU School of Chemistry and Chemical Engineering Henan University of Technology Zhengzhou 450052

Ionic liquids （ILs） as additive in phosphate buffer for detection of carbofuran by plant-lipases inhibition method is described. The higher inhibition efficiency and the shorter analysis time can be obtained by usin... 详细信息

关键词： Ionic liquid carbofuran plant-lipases.