带惩罚项与随机输入的BP神经网络在线梯度学习算法的收敛性

本文对三层BP神经网络中带有惩罚项的在线梯度学习算法的收敛性问题进行了研究,在网络训练每一轮开始执行之前,对训练样本随机进行重排,以使网络学习更容易跳出局部极小,文中给出了误差函数的单调性定理以及该算法的弱收敛和强收敛性定理。     

Dynamic Multicast Session Provisioning in WDM Optical Networks with Sparse Splitting Capability

In this work, we study dynamic provisioning of multicast sessions in a wavelength-routed sparse splitting capable WDM network with an arbitrary mesh topology where the network consists of nodes with full, partial, or no wavelength conversion capabilities and a node can be a tap-and-continue (TaC) node or a splitting and delivery (SaD) node. The objectives are to minimize the network resources in terms of wavelength-links used by each session and to reduce the multicast session blocking probability. The problem is to route the multicast session from each source to the members of every multicast session, and to assign an appropriate wavelength to each link used by the session. We propose an efficient online algorithm for dynamic multicast session provisioning. To evaluate the proposed algorithm, we apply the integer linear programming (ILP) optimization tool on a per multicast session basis to solve off-line the optimal routing and wavelength assignment given a multicast session and the current network topology as well as its residual network resource information. We formulate the per session multicast routing and wavelength assignment problem as an ILP. With this ILP formulation, the multicast session blocking probability or success probability can then be estimated based on solving a series of ILPs off-line. We have evaluated the effectiveness of the proposed online algorithm via simulation in terms of session blocking probability and network resources used by a session. Simulation results indicate that our proposed computationally efficient online algorithm performs well even when a fraction of the nodes are SaD nodes.     

基于进化人工神经网络的组合系统滤波技术

在组合系统运用Kalman滤波器技术时,准确的系统模型和可靠的观测数据是保证其性能的重要因素,否则将大大降低Kalman滤波器的估计精度,甚至导致滤波器发散.为解决上述Kalman应用中的实际问题,提出了一种新颖的基于进化人工神经网络技术的自适应Kalman滤波器.仿真试验表明该算法可以在系统模型不准确时、甚至外部观测数据短暂中断时,仍能保证Kalman滤波器的性能.     

应用于ROHC的窗基LSB编解码修正算法实现机制

在移动：IPv6中，常常因为数据分组的头标太大而消耗过多的带宽，因而需要使用灵活可靠的鲁棒性头标压缩机制(ROHC)来对分组头标进行压缩。在ROHC中，使用LSB编码方案对分组中那些变化比较小的头标域进行编码。本文将在简要介绍ROHC原理的基础上，深入探讨应用于ROHC的窗基LSB编解码修正算法的实现机制。     

TCP/IP网络的动态模型描述

拥塞现象成为TCP/IP网络发展面临的一个重要问题。因此，拥塞控制对TCP/IP网络的鲁棒性和稳定性具有重要作用。目前，网络拥塞控制策略主要包括两类：端到端的控制机制，如TCP拥塞控制算法；网络内部的主动队列管理(AQM)策略。但由于缺乏对网络系统动态特性的了解，这些拥塞控制策略大都基于专家经验，并没有建立完整的理论分析框架。为此，本文从数据流的角度出发，通过建立网络基本单元状态方程模型来实现IP网络系统的数学解析模型，然后用混杂系统来描述TCP带有拥塞控制策略的数据传输过程，建立了TCP/IP网络的动态模型，为网络系统中动态性能的分析、拥塞控制策略的设计奠定了基础。实验结果表明，该数学模型与NS仿真实验的结果相一致。     

透析3G在中国面临的重重困局(1)--牌照与标准选择

探讨了中国3G牌照何时发放、发放数量、牌照与标准是否捆绑、3G产业链如何搭建、频率如何分配、市场容量和终端等问题，并对此提出了建设性的意见。     

Survivability in IP-over-WDM Networks: WDM Lightpath Protection versus IP LSP Protection

This article compares the performance of WDM lightpath protection and IP LSP protection schemes for IP-over-WDM networks. A mathematical formulation of the maximum throughput problem is presented and analytical expressions for recovery time are derived for both schemes. The throughputs and recovery times are analyzed and compared. Results show that the IP LSP protection scheme presents higher throughputs then WDM lightpath protection. The IP LSP protection scheme, providing individual IP LSP protection has, however, scalability problems. This scheme presents high recovery times when a failure affects many lightpaths and many hops are allowed for the primary routes of IP LSPs.     

On the Existence of Solutions to Vector Quasivariational Inequalities and Quasicomplementarity Problems with Applications Break to Traffic Network Equilibria

For vector quasivariational inequalities involving multifunctions in topological vector spaces, an existence result is obtained without a monotonicity assumption and with a convergence assumption weaker than semicontinuity. A new type of quasivariational inequality is proposed. Applications to quasicomplementarity problems and traffic network equilibria are considered. In particular, definitions of weak and strong Wardrop equilibria are introduced for the case of multivalued cost functions.     

The Downs-Thomson Paradox: Existence, Uniqueness and Stability of User Equilibria

Consider a network where two routes are available for users wishing to travel from a source to a destination. On one route (which could be viewed as private transport) service slows as traffic increases. On the other (which could be viewed as public transport) the service frequency increases with demand. The Downs-Thomson paradox occurs when improvements in service produce an overall decline in performance as user equilibria adjust. Using the model proposed by Calvert [10], with a ⋅|M|1 queue corresponding to the private transport route, and a bulk-service infinite server queue modelling the public transport route, we give a complete analysis of this system in the setting of probabilistic routing. We obtain the user equilibria (which are not always unique), and determine their stability.AMS subject classification: 60K30, 90B15, 90B20, 91A10, 91A13This revised version was published online in June 2005 with corrected coverdate     

Parallel Algorithm for Unconstrained Optimization Based on Decomposition Techniques

We present a numerical implementation of the parallel gradient distribution (PGD) method for the solution of large-scale unconstrained optimization problems. The proposed parallel algorithm is characterized by a parallel phase which exploits the portions of the gradient of the objective function assigned to each processor; then, a coordination phase follows which, by a synchronous interaction scheme, optimizes over the partial results obtained by the parallel phase. The parallel and coordination phases are implemented using a quasi-Newton limited-memory BFGS approach. The computational experiments, carried out on a network of UNIX workstations by using the parallel software tool PVM, show that parallelization efficiency was problem dependent and ranged between 0.15 and 8.75. For the 150 problems solved by PGD on more than one processor, 85 cases had parallelization efficiency below 1, while 65 cases had a parallelization efficiency above 1.