Towards high performance low bitwidth training for deep neural networks

The high performance of the state-of-the-art deep neural networks(DNNs)is acquired at the cost of huge consumption of computing resources.Quantization of networks is recently recognized as a promising solution to solve the problem and significantly reduce the resource usage.However,the previous quantization works have mostly focused on the DNN inference,and there were very few works to address on the challenges of DNN training.In this paper,we leverage dynamic fixed-point(DFP)quantization algorithm and stochastic rounding(SR)strategy to develop a fully quantized 8-bit neural networks targeting low bitwidth training.The experiments show that,in comparison to the full-precision networks,the accuracy drop of our quantized convolutional neural networks(CNNs)can be less than 2%,even when applied to deep models evaluated on Image-Net dataset.Additionally,our 8-bit GNMT translation network can achieve almost identical BLEU to full-precision network.We further implement a prototype on FPGA and the synthesis shows that the low bitwidth training scheme can reduce the resource usage significantly.     

Topological properties and routing algorithm for semi-diagonal toms networks

To improve the scalability and reduce the implementation complexity of Mesh and Mesh-like networks,the semi-diagonal Torus(SD-Torus)network,a regular and symmetrical interconnection network is proposed.The SD-Torus network is a combination of a typical 2D-Torus network with two extra diagonal links from northwest to southeast direction for each node.The topological properties of SD-Torus networks are discussed,and a load balanced routing algorithm for SD-Torus is presented.System-C based simulation result shows that,compared with diagonal Mesh(DMesh),diagonal Torus(DTorus)and XMesh networks,the SD-Torus network can achieve high performance with a lower network cost.It makes the SD-Torus network a powerful candidate for the high performance interconnection networks.     

Mobile Internet Application Infrastructure: An Open Garden

Based on good mobile Internet application infrastructure, the user access layer can provide users with quick and easy access to mobile Internet applications; the service provisioning layer can improve both access rate and Quality of Service (QoS), enforce the control of content resources, and reduce the cost for content storage as well as management overhead; and the service layer can reduce service development requirements and operation and maintenance cost, making it easier to launch services. This paper suggests that the Chinese Government should, based on the existing networks, make strategic efforts to push research work and industrialization process of the mobile Internet application infrastructure.     

混合无线Mesh网络中基于链路层协作的路由方案(英文)

In a Wireless Mesh Network(WMN),the convenience of a routing strategy strongly depends on the mobility of the intermediate nodes that compose the paths.Taking this behaviour into account,this paper presents a routing scheme that works differently accordingly to the node mobility.In this sense,a proactive routing scheme is restricted to the backbone to promote the use of stable routes.Conversely,the reactive protocol is used for searching routes to or from a mobile destination.Both approaches are simultaneously implemented in the mesh nodes so that the routing protocols share routing information that optimises the network performance.Aimed at guaranteeing the IP compatibility,the combination of the two protocols in the core routers is carried out in the Medium Access Control(MAC)layer.In contrast to the operation in the IP layer where two routing protocols cannot work concurrently,the transfer of the routing tasks to the MAC layer enables the use of multiple independent forwarding tables.Simulation results show the advantage of the proposal in terms of packet losses and data delay.     

A Game-Theoretic Perspective on Resource Management for Large-Scale UAV Communication Networks

As a result of rapid development in electronics and communication technology,large-scale unmanned aerial vehicles(UAVs)are harnessed for various promising applications in a coordinated manner.Although it poses numerous advantages,resource management among various domains in large-scale UAV communication networks is the key challenge to be solved urgently.Specifically,due to the inherent requirements and future development trend,distributed resource management is suitable.In this article,we investigate the resource management problem for large-scale UAV communication networks from game-theoretic perspective which are exactly coincident with the distributed and autonomous manner.By exploring the inherent features,the distinctive challenges are discussed.Then,we explore several gametheoretic models that not only combat the challenges but also have broad application prospects.We provide the basics of each game-theoretic model and discuss the potential applications for resource management in large-scale UAV communication networks.Specifically,mean-field game,graphical game,Stackelberg game,coalition game and potential game are included.After that,we propose two innovative case studies to highlight the feasibility of such novel game-theoretic models.Finally,we give some future research directions to shed light on future opportunities and applications.     

Internet Plus and Networks Convergence

This paper discusses the relationship between Internet plus and networks convergence and highlights the development of ICT industry and the process of networks convergence. Based on Internet plus and the theory of networks convergence, this paper suggests to implement Internet plus action plan as a strategy to accelerate the construction of networks and application of Internet, promote the t raditional industry reform with ICT, increase competitive efficiency, standardize the market competition, and then realize the Internet plus strategy and networks convergence.     

Space-Air-Ground Integrated Network with Native Intelligence (NI-SAGIN):Concept,Architecture,Access and Networking

Ubiquitous coverage is one of the most important goals for mobile communication networks.To achieve this,integration of space,air,and ground networks is highly demanded,which expects to become the one of the enabling technologies for 6G networks.     

MAC Layer Resource Allocation for Wireless Body Area Networks

Wireless body area networks(WBANs) can provide low-cost, timely healthcare services and are expected to be widely used for ehealthcare in hospitals. In a hospital, space is often limited and multiple WBANs have to coexist in an area and share the same channel in order to provide healthcare services to different patients. This causes severe interference between WBANs that could significantly reduce the network throughput and increase the amount of power consumed by sensors placed on the body. Therefore, an efficient channel-resource allocation scheme in the medium access control(MAC) layer is crucial. In this paper, we develop a centralized MAC layer resource allocation scheme for a WBAN. We focus on mitigating the interference between WBANs and reducing the power consumed by sensors. Channel and buffer state are reported by smartphones deployed in each WBAN,and channel access allocation is performed by a central controller to maximize network throughput. Sensors have strict limitations in terms of energy consumption and computing capability and cannot provide all the necessary information for channel allocation in a timely manner. This deteriorates network performance. We exploit the temporal correlation of the body area channel in order to minimize the number of channel state reports necessary. We view the network design as a partly observable optimization problem and develop a myopic policy, which we then simulate in Matlab.     

Performance analysis and optimization model based on dynamics equation for heterogeneous wireless networks

In this paper, we propose a model based on dynamics equation for performance analysis and optimization for heterogeneous wireless networks(HWNs). First, the channel occupation state with time of HWNs is modelled by dynamics equation, in which users’ mobility as an important factor affecting system performance is considered. Then the steady state probability distribution of channel occupation is derived. Based on the results, the expression of the throughput of HWNs is deduced, which includes a factor ρ which is the ratio of service arrival rate accessing one of the networks to the total service arrival rate in the overlapping area. And this paper proposes to maximize the throughput of the HWNs by optimizing the factor ρ to efficiently utilize the resources. Simulation results show that the proposed optimization method can effectively improve the throughput and in the meanwhile decrease the blocking probability of the whole system.     

Advanced Leader Election for Virtual Traffic Lights

We examine the network performance of algorithms for self-organized traffic management. In particular, we focus on wireless network-ing between cars. One of many technologies that make road traffic safer and more efficient is the Virtual Traffic Light(VTL) system,which is able to coordinate the traffic flow at intersections without the need for physical lights. VTL takes a leading vehicle at an inter-section and uses it to control the traffic lights. We developed algorithms for leader election and traffic light computation in realistic ve-hicular networking scenarios. Our key contribution is the extension of this algorithm to support arbitrary intersection layouts. We in-vestigated the proposal in synthetic and realistic scenarios. The results show that, overall, VTLs use network resources efficiently andpositively influences driving experience. It performs better than stationary traffic lights for a low to medium network load. We alsoidentify potential optimizations to deal with high network load and to improve fairness.     

Cross-layer congestion control in ad hoc wireless networks

The paper presents the problem of performance degradation of transport layer protocols due to congestion of wireless local area networks. Following the analysis of available solutions to this problem, a cross-layer congestion avoidance scheme (C³TCP) is presented, able to obtain higher performance by gathering capacity information such as bandwidth and delay at the link layer. The method requires the introduction of an additional module within the protocol stack of the mobile node, able to adjust the outgoing data stream based on capacity measurements. Moreover, a proposal to provide optional field support to existing IEEE 802.11 protocol, in order to support the presented congestion control solution as well as many other similar approaches, is presented. Achieved results underline good agreement with design considerations and high utilization of the available resources.     

Application versus network layer multicasting in ad hoc networks: the ALMA routing protocol

Application layer multicasting has emerged as an appealing alternative to network layer multicasting in wireline networks. Here, we examine the suitability of application layer multicast in ad hoc networks. To this effect, we propose a flexible receiver-driven overlay multicast protocol which we call Application Layer Multicast Algorithm (ALMA). ALMA constructs an overlay multicast tree in a dynamic, decentralized and incremental way. First, ALMA is receiver-driven: the member nodes find their connections according to their needs. Second, it is flexible, and thus, it can satisfy the performance goals and the needs of a wide range of applications. Third, it is highly adaptive: it reconfigures the tree in response to mobility or congestion. In addition, our protocol has the advantages of an application layer protocol: (a) simplicity of deployment, (b) independence from lower layer protocols, and (c) capability of exploiting features such as reliability and security that may be provided by the lower layers. Through extensive simulations, we show that ALMA performs favorably against the currently best application layer and network layer protocols. In more detail, we find that ALMA performs significantly better than ODMRP, a network layer, for small group sizes. We conclude that the application layer approach and ALMA seem very promising for ad hoc multicasting.     

A survey of free space optical networks

Free Space Optical (FSO) networks, also known as optical wireless networks, have emerged as viable candidates for broadband wireless communications in the near future. The range of the potential application of FSO networks is extensive, from home to satellite. However, FSO networks have not been popularized because of insufficient availability and reliability. Researchers have focused on the problems in the physical layer in order to exploit the properties of wireless optical channels. However, recent technological developments with successful results make it practical to explore the advantages of the high bandwidth. Some researchers have begun to focus on the problems of network and upper layers in FSO networks. In this survey, we classify prospective global FSO networks into three subnetworks and give an account of them. We also present state-of-the-art research and discuss what kinds of challenges exist.     

基于FPGA的汽车电子设计

随着汽车电子产业的快速发展．传统MCU、ASIC已很难满足汽车电子设计灵活性、稳定性、低成本、快速开发等要求,提出一种基于FPGA的汽车电子设计方案。介绍FPGA的特点及其在汽车电子设计中的应用,并介绍一款基于NiosⅡ软核处理器。用SoPC技术设计的具有密码锁及GSM／GPS防盗系统的汽车电子系统,FPGA将成为未来汽车电子设计的理想选择,     

基于FPGA的万兆网的IPsec ESP协议设计与实现

“Internet协议安全性(IPsec)”为IP层及其上层协议提供加解密和认证等安全服务。但对IPsec协议的处理已经成为高速网络实现的瓶颈。随着FPGA向着更大容量和更高速度方向发展,基于FPGA硬件实现的IPsec协议栈可以提供更高的网络性能。文中介绍了一种基于FPGA的万兆以太网IPsec ESP协议栈的设计,支持隧道模式和传输模式,具有抗重放能力。通过采用多级流水操作、多缓存乒乓操作、多进程并行处理等技术实现了万兆线速。     

基于FPGA的汽车电子设计

随着汽车电子产业的快速发展,传统MCU、ASIC已很难满足汽车电子设计灵活性、稳定性、低成本、快速开发等要求.提出一种基于FPGA的汽车电子设计方案.介绍FPGA的特点及其在汽车电子设计中的应用,并介绍一款基于Nios II软核处理器.用SoPC技术设计的具有密码锁及GSM/GPS防盗系统的汽车电子系统,FPGA将成为未来汽车电子设计的理想选择.     

基于FPGA的ATM层与适配层的设计与实现

文中介绍了ATM网络各层工作原理和FPGA的结构及工作原理．结合ATM网络各层的处理过程，重点阐述了ATM网络各层在FPGA中的硬件结构及实现，提出了一种基于FPGA的ATM层和适配层处理的实现方案．     

FPGA-based acceleration for binary neural networks in edge computing

As a core component in intelligent edge computing, deep neural networks (DNNs) will increasingly play a critically important role in addressing the intelligence-related issues in the industry domain, like smart factories and autonomous driving. Due to the requirement for a large amount of storage space and computing resources, DNNs are unfavorable for resource-constrained edge computing devices, especially for mobile terminals with scarce energy supply. Binarization of DNN has become a promising technology to achieve a high performance with low resource consumption in edge computing. Field-programmable gate array (FPGA)-based acceleration can further improve the computation efficiency to several times higher compared with the central processing unit (CPU) and graphics processing unit (GPU). This paper gives a brief overview of binary neural networks (BNNs) and the corresponding hardware accelerator designs on edge computing environments, and analyzes some significant studies in detail. The performances of some methods are evaluated through the experiment results, and the latest binarization technologies and hardware acceleration methods are tracked. We first give the background of designing BNNs and present the typical types of BNNs. The FPGA implementation technologies of BNNs are then reviewed. Detailed comparison with experimental evaluation on typical BNNs and their FPGA implementation is further conducted. Finally, certain interesting directions are also illustrated as future work.     

基于覆盖网络的多路径流量均衡研究

现有网络发送数据都集中在一条最优路径上,使得网络流量不均衡,拥塞不可避免.传统的IP网络难以拓展,文章在应用层上对覆盖网络进行流量均衡,利用覆盖网络之间的互动性给出了覆盖网络的多路径流量均衡模型,在该模型上研究了输入流量和输出流量的分配算法,并通过仿真实验说明了可以有效地减小网络的端到端延迟.