Towards efficient deep neural network training by FPGA-based batch-level parallelism

Training deep neural networks(DNNs)requires a significant amount of time and resources to obtain acceptable results,which severely limits its deployment in resource-limited platforms.This paper proposes DarkFPGA,a novel customizable framework to efficiently accelerate the entire DNN training on a single FPGA platform.First,we explore batch-level parallelism to enable efficient FPGA-based DNN training.Second,we devise a novel hardware architecture optimised by a batch-oriented data pattern and tiling techniques to effectively exploit parallelism.Moreover,an analytical model is developed to determine the optimal design parameters for the DarkFPGA accelerator with respect to a specific network specification and FPGA resource constraints.Our results show that the accelerator is able to perform about 10 times faster than CPU training and about a third of the energy consumption than GPU training using 8-bit integers for training VGG-like networks on the CIFAR dataset for the Maxeler MAX5 platform.

Artificial Neural Network Modeling of Healthy Risk Level Induced by Aircraft Pollutant Impacts around Soekarno Hatta International Airport

Aircraft pollutant emissions are an important part of sources of pollution that directly or indirectly affect human health and ecosystems. This research suggests an Artificial Neural Network model to determine the healthy risk level around Soekarno Hatta International Airport-Cengkareng Indonesia. This ANN modeling is a flexible method, which enables to recognize highly complex non-linear correlations. The network was trained with real measurement data and updated with new measurements, enhancing its quality and making it the ideal method for this research. Measurements of aircraft pollutant emissions are carried out with the aim to be used as input data and to validate the developed model. The obtained results concerned the improved ANN architecture model based on pollutant emissions as input variables. ANN model processes variables—hidden layers—and gives an output variable corresponding to a healthy risk level. This model is characterized by a 4-10-1 scheme. Based on ANN criteria, the best validation performance is achieved at epoch 28 from 34 epochs with the Mean Squared Error (MSE) of 9 × 10-3. The correlation between targets and outputs is confirmed. It validated a close relationship between targets and outputs. The network output errors value approaches zero. Further research is needed with the aim to enlarge the scheme of the ANN model by increasing its input variables. This is one of the major key defining environmental capacities of an airport that should be applied by Indonesian airport authorities. These would institute policies to manage or reduce pollutant emissions considering population and income growth to be socially positive.

Water level updating model for flow calculation of river networks

Complex water movement and insufficient observation stations are the unfavorable factors in improving the accuracy of flow calculation of river networks. A water level updating model for river networks was set up based on a three-step method at key nodes, and model correction values were collected from gauge stations. To improve the accuracy of water level and discharge forecasts for the entire network, the discrete coefficients of the Saint-Venant equations for river sections were regarded as the media carrying the correction values from observation locations to other cross-sections of the river network system. To examine the applicability, the updating model was applied to flow calculation of an ideal river network and the Chengtong section of the Yangtze River. Comparison of the forecast results with the observed data demonstrates that this updating model can improve the forecast accuracy in both ideal and real river networks.

Application of Neural network PID Controller in Constant Temper-ature and Constant Liquid-level System

Guided by the principle of neural network, an intelligent PID controller based on neural network is devised and applied to control of constant temperature and constant liquidlevel system. The experiment results show that this controller has high accuracy and strong robustness and good characters.

Network Traffic Generation Based on Statistical Packet-Level Characteristics

Network traffic is very important for testing network equipment, network services, and security products. A new method of generating traffic based on statistical packet-level characteristics is proposed. In every time unit, the generator determines the sent packets number, the type and size of every sent packet according to the statistical characteristics of the original traffic. Then every packet, in which the protocol headers of transport layer, network layer and ethernet layer are encapsulated, is sent via the responding network interface card in the time unit. The results in the experiment show that the correlation coefficients between the bandwidth, the packet number, packet size distribution, the fragment number of the generated network traffic and those of the original traffic are all more than 0.96. The generated traffic and original traffic are very highly related and similar.

CMLP: Exploiting Caches at Multiple Levels of Proxies to Enhance Seamless Mobility Support in Information-Centric Networks

The recent evolution of the Internet towards "Information-centric" transfer modes has renewed the interest in exploiting proxies to enhance seamless mobility. In this work, we focus on the case of multiple levels of proxies in ICN architectures, in which content requests from mobile subscribers and the corresponding items are proactively cached to these proxies at different levels. Specifically, we present a multiple-level proactive caching model that selects the appropriate subset of proxies at different levels and supports distributed online decision procedures in terms of the tradeoff between delay and cache cost. We show via extensive simulations the reduction of up to 31.63% in the total cost relative to Full Caching, in which caching in all 1-level neighbor proxies is performed, and up to 84.21% relative to No Caching, in which no caching is used. Moreover, the proposed model outperforms other approaches with a flat cache structure in terms of the total cost.

Analysis of Lifetime of Large Wireless Sensor Networks Based on Multiple Battery Levels

Due to the limited transmission range, data sensed by each sensor has to be forwarded in a multi-hop fashion before being delivered to the sink. The sensors closer to the sink have to forward comparatively more messages than sensors at the periphery of the network,and will deplete their batteries earlier. Besides the loss of the sensing capabilities of the nodes close to the sink, a more serious consequence of the death of the first tier of sensor nodes is the loss of connectivity between the nodes at the periphery of the network and the sink;it makes the wireless networks expire. To alleviate this undesired effect and maximize the useful lifetime of the network, we investigate the energy consumption of different tiers and the effect of multiple battery levels, and demonstrate an attractively simple scheme to redistribute the total energy budget in multiple battery levels by data traffic load. We show by theoretical analysis, as well as simulation, that this substantially improves the network lifetime.

Research on confirmation of basic technological parameters of tension levellers based on neural network and genetic algorithm

Confirmation of basic technological parameters of tension levellers is the most important factor of leveling strip. Up to now, most factories have used experts’ experience to decide these parameters, without any established rule to follow. For better quality of strip, a valid method is needed to decide technological parameters precisely and reasonably. In this paper, a method is used based on neural network and genetic algorithm. Neural network has a good ability to extract rules from work process of tension levellers. Then using neural network, which has learned from a lot of working samples, to be the evaluation of fitness, genetic algorithm could easily find the best or better technological parameters. At the end of this paper, examinations are given to show the effect of this method.

Robust Synchronization in an E/I Network with Medium Synaptic Delay and High Level of Heterogeneity

It is known that both excitatory and inhibitory neuronal networks can achieve robust synchronization only under certain conditions, such as long synaptic delay or low level of heterogeneity. In this work, robust synchronization can be found in an excitatory/inhibitory （E/I） neuronal network with medium synaptie delay and high level of heterogeneity, which often occurs in real neuronal networks. Two effects of post-synaptic potentials （PSP） to network synchronization are presented, and the synaptic contribution of excitatory and inhibitory neurons to robust synchronization in this E/I network is investigated. It is found that both excitatory and inhibitory neurons may contribute to robust synchronization in E/I networks, especially the excitatory PSP has a more positive effect on synchronization in E/I networks than that in excitatory networks. This may explain the strong robustness of synchronization in Eli neuronal networks.

Parameter Method Data Processing for CPⅢ Precise Trigonometric Leveling Network

In view of the limitation of the difference method,the adjustment model of CPⅢprecise trigonometric leveling control network based on the parameter method was proposed in the present paper.The experiment results show that this model has a simple algorithm and high data utilization,avoids the negative influences caused by the correlation among the data acquired from the difference method and its accuracy is improved compared with the difference method.In addition,the strict weight of CPⅢprecise trigonometric leveling control network was also discussed in this paper.The results demonstrate that the ranging error of trigonometric leveling can be neglected when the vertical angle is less than 3 degrees.The accuracy of CPⅢprecise trigonometric leveling control network has not changed significantly before and after strict weight.

High-Level Portable Programming Language for Optimized Memory Use of Network Processors

Network processors (NPs) are widely used for programmable and high-performance networks;however, the programs for NPs are less portable, the number of NP program developers is small, and the development cost is high. To solve these problems, this paper proposes an open, high-level, and portable programming language called “Phonepl”, which is independent from vendor-specific proprietary hardware and software but can be translated into an NP program with high performance especially in the memory use. A common NP hardware feature is that a whole packet is stored in DRAM, but the header is cached in SRAM. Phonepl has a hardware-independent abstraction of this feature so that it allows programmers mostly unconscious of this hardware feature. To implement the abstraction, four representations of packet data type that cover all the packet operations (including substring, concatenation, input, and output) are introduced. Phonepl have been implemented on Octeon NPs used in plug-ins for a network-virtualization environment called the VNode Infrastructure, and several packet-handling programs were evaluated. As for the evaluation result, the conversion throughput is close to the wire rate, i.e., 10 Gbps, and no packet loss (by cache miss) occurs when the packet size is 256 bytes or larger.

Neural Network Based Multi-level Fuzzy Evaluation Model for Mechanical Kinematic Scheme

To implement a quantificational evaluation for mechanical kinematic scheme more effectively,a multi-level and multi-objective evaluation model is presented using neural network and fuzzy theory. Firstly,the structure of evaluation model is constructed according to evaluation indicator system. Then evaluation samples are generated and provided to train this model. Thus it can reflect the relation between attributive value and evaluation result,as well as the weight of evaluation indicator. Once evaluation indicators of each candidate are fuzzily quantified and fed into the trained network model,the corresponding evaluation result is outputted and the best alternative can be selected. Under this model,expert knowledge can be effectively acquired and expressed,and the quantificational evaluation can be implemented for kinematic scheme with multi-level evaluation indicator system. Several key problems on this model are discussed and an illustration has demonstrated that this model is feasible and can be regarded as a new idea for solving kinematic scheme evaluation.

Groundwater Level Prediction Using Artificial Neural Networks: A Case Study in Tra Noc Industrial Zone, Can Tho City, Vietnam

The objective of this study is to predict groundwater levels (GWLs) under different impact factors using Artificial Neural Network (ANN) for a case study in Tra Noc Industrial Zone, Can Tho City, Vietnam. This can be achieved by evaluating the current state of groundwater resources (GWR) exploitation, use and dynamics;setting-up, calibrating and validating the ANN;and then predicting GWLs at different lead times. The results show that GWLs in the study area have been found to reduce rapidly from 2000 to 2015, especially in the Middle-upper Pleistocene (qp2-3) and upper Pleistocene (qp3) due to the over-withdrawals from the enterprises for production purposes. Concerning this problem, an Official Letter of the People’s Committee of Can Tho City was issued and taken into enforcement in 2012 resulting in the reduction of exploitation. The calibrated ANN structures have successfully demonstrated that the GWLs can be predicted considering different impact factors. The predicted results will help to raise awareness and to draw an attention of the local/central government for a clear GWR management policy for the Mekong delta, especially the industrial zones in the urban areas such as Can Tho city.

考虑大规模风光分层接入的配电网多层协调无功优化方法

大量风光新能源分层、多点接入给配电网运行带来了电压越限等负面影响,增加了不同电压等级无功协同优化的难度。兼顾不同电压等级调节需求,从模型降维等值的角度,提出了基于等值模型的配电网多层协调无功优化方法。该方法分别针对含有无功调节设备及没有无功调节设备的两种台区系统,利用神经网络对大量台区相关运行数据进行训练得到台区拟合模型,形成不可控和可控两种类型的台区等值模型,并用这两类拟合模型分别代替两类台区系统的物理模型,使台区以数据驱动的方式参与配电网无功优化,形成馈线物理模型和台区拟合模型组成的单一电压等级物数混合优化调度模型。将原多层无功优化问题转换为单层系统优化问题,再对该单层系统无功优化调度问题进行求解。该方法可降低配电网系统优化计算规模,从而减小计算量,以无功优化数学模型与数据驱动方法相结合的方式,实现配电网馈线-台区多层协调无功优化。通过算例验证了所提方法对于解决分布式风光分层接入所导致电压越限问题的可行性、有效性和优越性,能够保障配电网的经济安全稳定运行。

基于分布鲁棒优化的车-站-网日前能量管理与交易

为考虑上级电网电价、光伏出力等多重多层级不确定性对车-站-网互动博弈模型的影响,且充分体现配电网主动管理技术支撑效果,文中提出了一种基于分布鲁棒优化的车-站-网能量管理与交易方法。首先,针对主动配电网内多元主体能量管理与交易问题,建立了配电网运营商、充电站和电动汽车的日前市场互动框架。其次,融合主动网络管理技术和网络约束,在配电网运营商与聚合了电动汽车的多个充电站之间构建了以多主体各自利益最大为目标的双层Wasserstein分布鲁棒互动博弈模型。然后,提出了结合Karush-Kuhn-Tucker条件、对偶原理和大M法的化简方法以解决多层级不确定性造成的求解难题,将双层Wasserstein分布鲁棒模型转化为单层混合整数二阶锥规划模型,并利用商业求解器YALMIP/GUROBI进行了求解。最后,通过算例仿真验证了所提模型和方法的有效性。

地铁网络中断下韧性评估及修复策略

随着我国城市地铁发展,地铁系统在面对自然灾害、突发事件等扰动事件时所表现出的韧性是目前备受关注的问题。为评估和优化地铁网络韧性,首先,深入分析拓扑结构,提出一种综合考虑网络连通水平和冗余水平的网络性能函数,并构建网络韧性评价指标;然后,在中断场景下,研究以韧性最优为目标的车站修复调度问题,采用遗传算法求解模型,明确各施工队的施工方案;最后,以台北捷运为例进行实例研究,验证方法的有效性。结果表明:多车站失效后,网络连通水平和冗余水平均有所下降,网络性能随之降低;在工期等约束下,最佳修复策略模型能够合理地制定施工方案以最大限度地提高地铁网络韧性;条件允许情况下,适当延长工期有助于进一步地提升最佳网络韧性。案例研究为城市地铁线网规划及提高地铁网络韧性提供有益借鉴。

基于多粒度阅读器和图注意力网络的文档级事件抽取

文档级事件抽取面临论元分散和多事件两大挑战,已有工作大多采用逐句抽取候选论元的方式,难以建模跨句的上下文信息。为此,提出了一种基于多粒度阅读器和图注意网络的文档级事件抽取模型,采用多粒度阅读器实现多层次语义编码,通过图注意力网络捕获实体对之间的局部和全局关系,构建基于实体对相似度的剪枝完全图作为伪触发器,全面捕捉文档中的事件和论元。在公共数据集ChFinAnn和DuEE-Fin上进行了实验,结果表明提出的方法改善了论元分散问题,提升了模型事件抽取性能。

国家水网各层级协同建设思考

推进国家骨干网和省市县各层级水网互联互通、合理衔接,对逐步形成国家水网“一张网”、更好发挥水网整体效益意义重大。推进各层级水网协同建设,应聚焦“四个坚持”原则,即坚持目标导向和问题导向相结合,坚持遵循自然规律、经济规律、生态规律,坚持系统观念,坚持“四水四定”;做好“五个统筹”,即统筹好水网布局、结构、功能设计方案与组网方案,统筹好不同层级水网规划布局与建设,统筹好资源环境承载能力与水流调配过程,统筹好骨干工程与配套工程的关系,统筹好常态与非常态协同治理。分析了不同层级水网协同建设的总体思路,从水网规划与建设的基本单元、思路与重点等方面,研究提出了不同层级水网协同建设的侧重点。围绕加快构建国家水网主骨架和大动脉、加强国家骨干网和省级水网互联互通、推进省市县水网协同融合、加强战略储备水源和战略输配水通道建设,提出推进各层级水网协同建设的重点举措,从完善国家水网“一张网”顶层设计、加强体制机制创新、强化资源要素保障等方面,研究提出了不同层级水网协同建设的政策建议。

Malicious Node Detection Using Confidence Level Evaluation in a Grid-Based Wireless Sensor Network

In this paper, we present a malicious node detection scheme using confidence-level evaluation in a grid-based wireless sensor network. The sensor field is divided into square grids, where sensor nodes in each grid form a cluster with a cluster head. Each cluster head maintains the confidence levels of its member nodes based on their readings and reflects them in decision-making. Two thresholds are used to distinguish between false alarms due to malicious nodes and events. In addition, the center of an event region is estimated, if necessary, to enhance the event and malicious node detection accuracy. Experimental results show that the scheme can achieve high malicious node detection accuracy without sacrificing normal sensor nodes.