Dynamics of learning and generalization in perceptrons with constraints

Depending on the kind of constraints imposed on the weights of a perceptron, learning can be a combinatorially hard problem. As an example of this type, I discuss a perception with binary weights comparing results obtained from replica theory, dynamic mean field theory and simulated annealing. Contrary to the replica calculation, dynamics yields information about the performance of a polynomial algorithm in a situation where the best solution cannot be found in polynomial time. I also discuss improved learning algorithms and results for finite size perceptrons.     

网络竞价系统中负载均衡技术优化

在网络系统正常运行中,如何确保数据准确实时高效的传输变得越来越重要,在一些高并发、高吞吐的服务中,确保每个事务功能正常运转越来越重要;系统的负载是否均衡是确保整个系统能否完整工作的关键,现在有很多负载均衡算法旨在使得每个服务器的负载趋于相等,可以归纳为静态均衡方法和动态均衡方法;考虑到影响服务器的负载有多重因数,结合贪心算法作出相对合理的负载均衡策略,应用到本次竞价系统中对比以前的平衡策略,通过比较可以得出多衡量指标的综合考虑会提高负载均衡的性能,对网络竞价系统能够起到更优化的作用。     

面向SLA的云计算负载均衡策略

云计算负载均衡是保障SLA协议的关键问题之一。针对云计算负载均衡问题,提出一种面向SLA的负载均衡策略。该策略引入人工神经网络思想,建立负载均衡模型,采用单层感知器算法(SLPA)将虚拟机负载状态进行分类,然后利用结合了动态加权轮询算法的BP神经网络算法(BPNNA-DWRRA)有针对性地对虚拟机负载权重进行预测更新,最后将任务调度到最小权重所对应的可行虚拟机上。应用CloudSim进行仿真实验,结果表明了该策略的可行性,同时,相比加权最小链接算法和粒子群算法,该策略的平均响应时间分别节省了43.6%和22.5%,SLA违反率分别降低了20.7%和14.4%。因此,所提策略在响应用户任务时,请求响应时间短,SLA违反率低,保障了SLA。     

Dynamics and stabilization of peak current-mode controlled buck converter with constant current load

The discrete iterative map model of peak current-mode controlled buck converter with constant current load(CCL),containing the output voltage feedback and ramp compensation, is established in this paper. Based on this model the complex dynamics of this converter is investigated by analyzing bifurcation diagrams and the Lyapunov exponent spectrum. The effects of ramp compensation and output voltage feedback on the stability of the converter are investigated. Experimental results verify the simulation and theoretical analysis. The stability boundary and chaos boundary are obtained under the theoretical conditions of period-doubling bifurcation and border collision. It is found that there are four operation regions in the peak current-mode controlled buck converter with CCL due to period-doubling bifurcation and border-collision bifurcation. Research results indicate that ramp compensation can extend the stable operation range and transfer the operating mode, and output voltage feedback can eventually eliminate the coexisting fast-slow scale instability.     

改进粒子群算法在云计算负载均衡中的应用研究

云计算系统采用虚拟化技术可以更加灵活和高效地分配运算资源,便于管理员根据用户任务需求按需分配云计算资源。但虚拟化后的云计算中心存在种类多样、数量庞大的虚拟机资源,难以将虚拟机合理地放置到物理主机集群上并达到较好的负载均衡。为此,给出了云计算中心虚拟机放置到物理主机的负载均衡模型,采用改进后的粒子群算法（PSO）来求解最优解。最后通过和常用虚拟机放置算法的仿真对比实验,验证了所提云计算负载均衡优化算法的有效性。     

MDSLB:A new static load balancing method for parallel molecular dynamics simulations

Large-scale parallelization of molecular dynamics simulations is facing challenges which seriously affect the simula- tion efficiency, among which the load imbalance problem is the most critical. In this paper, we propose, a new molecular dynamics static load balancing method （MDSLB）. By analyzing the characteristics of the short-range force of molecular dynamics programs running in parallel, we divide the short-range force into three kinds of force models, and then pack- age the computations of each force model into many tiny computational units called ＂cell loads＂, which provide the basic data structures for our load balancing method. In MDSLB, the spatial region is separated into sub-regions called ＂local domains＂, and the cell loads of each local domain are allocated to every processor in turn. Compared with the dynamic load balancing method, MDSLB can guarantee load balance by executing the algorithm only once at program startup without migrating the loads dynamically. We implement MDSLB in OpenFOAM software and test it on TianHe-lA supercomputer with 16 to 512 processors. Experimental results show that MDSLB can save 34%-64% time for the load imbalanced cases.     

Dynamic global load balancing strategy for cross stratum optimization of OpenFlow-enabled triple-M optical networks

With the emergence of high-bitrate applications, cross stratum optimization (CSO) attracts the interest of network operators because of its application in the joint optimization of optical networks and application stratum resources. Given the large-scale growth and high complexity of optical networks, achieving a more effective, accurate, and practical CSO becomes an important research focus. In this letter, we present a CSO-oriented, unified control architecture for OpenFlow-enabled triple-M optical networks. A novel dynamic global load balancing (DGLB) strategy with dynamic resource rating for CSO is presented based on the proposed architecture. The DGLB strategy is then compared with four other strategies by conducting experiments on a SOFT-based testbed with 1000 virtual nodes.     

Dynamics of an axially functionally graded beam under axial load

This paper considers the dynamics of a simply supported beam under axial time–dependent load. The beam is made of an axially functionally graded material. The motion equations are deduced from the equilibrium in deformed configuration and no restriction is made on the amplitude of the transversal displacement, but that naturally imposed by the inextensibility assumption that is adopted in the present study. The transversal motion equation, that is a partial differential equation, is approximated by its Taylor expansion until third order and then discretized through the Galerkin procedure.     

Dynamics of load entropy during cascading failure propagation in scale-free networks

In this Letter, we introduce the concept of load entropy, which can be an average measure of a network's heterogeneity in the load distribution. Then we investigate the dynamics of load entropy during failure propagation using a new cascading failures load model, which can represent the node removal mechanism in many real-life complex systems. Simulation results show that in the early stage of failure propagation the load entropy for a larger cascading failure increases more sharply than that for a smaller one, and consequently the cascading failure with a larger damage can be identified at the early stage of failure propagation according to the load entropy. Particularly, load entropy can be used as an index to be optimized in cascading failures control and defense in many real-life complex networks.     

Brownian dynamics with constraints

Systems possessing degrees of freedom operating on widely separated timescales, where the effects of those operating on the smaller timescales are relatively unimportant, may be modelled by the use of the Langevin equation. In order to study such systems containing complex polyatomic particles, holonomic constraints may be used. Though there is no lack of published algorithms for the numerical solution of the Langevin equation, few of them have been developed with sufficient rigour to ensure their precision, nor to demonstrate their compatibility with constraints. This study recapitulates an approach based upon Runge-Kutta equations which has the advantage of being perfectible to any desired order in the time-step, and shows how it may be combined with the SHAKE method in order to perform constrained Brownian dynamics simulations. Results are presented for some simple systems with a third order algorithm, and it is found that the correct dynamic and statistical behaviour is recovered.     

Mechanical systems with nonlinear constraints

A geometrical formalism for nonlinear nonholonomic Lagrangian systems is developed. The solution of the constrained problem is discussed by using almost product structures along the constraint submanifold. Constrained systems with ideal constraints are also considered, and Chetaev conditions are given in geometrical terms. A Noether theorem is also proved.     

Self-avoiding walks with geometrical constraints

We consider self-avoiding walks on aD-dimensional hypercubic lattice, confined to a slab geometry and confined to a half-space. We present a proof of the existence of a connective constant for the slab geometry and review some corresponding results for the half-space. We also discuss the way in which scaling arguments can be used to give stronger, but nonrigorous, results.Presented at the Symposium on Random Walks, Gaithersburg, MD, June 1982.     

Quantization of relativistic systems with constraints

The general dynamical system with constraints is quantized, and the S-matrix is constructed in the most general class of gauges including relativistic ones. In the case when constraints do not form a group a new type of additional diagrams arises securing unitarity of the theory: the four-fermion interaction of ghost fields.     

Higher-order mechanical systems with constraints

We extend Weber's approach for mechanical systems with constraints in terms of differential-geometric structures underlying higher-order tangent bundles.     

Canonical quantization with quadratic constraints

Based on an analysis first suggested by Bryce S. DeWitt, we have found that a special case of the general classical theory involving quadratic constraints can be quantized canonically, in the sense that the quantum constraints are consistent. In particular, this special case contains all known physical theories of bosons,including Einstein'sGeneral Theory of Relativity. The quantum constraints for this theory are given explicitly in an appendix.     

Stochastic variational problems with constraints

Two theorems in stochastic variational problems with constraints are presented.Partially supported by the Swiss National Science Foundation.     

Quantum Hamiltonian formalism with constraints

A quantum system with constraints that does not necessarily correspond to a classical system with constraints is described in the Hamiltonian formalism.     

Modal balancing of flexible rotors with bow and distributed unbalance

Unbalance and bow are found to be one of the most common causes of synchronous machinery vibrations in rotating systems. Concentrated lumped mass models are adopted in most of the finite element approach for modeling unbalances and subsequent balancing in rotating systems. But this assumption may not be appropriate for long slender rotors with unbalances distributed along the length of the rotor. A polynomial curve for eccentricity distribution with finite element modeling is used to identify the distributed unbalance. The unbalance eccentricity distributions are estimated using the measured vibration responses at a speed below the balancing speed. Modal correction mass required to balance a rotor at its first bending critical speed, having both distributed unbalance and bow is computed knowing the amplification factor at critical speed. The rotor is balanced at its first bending critical speed using modal balancing method in a single trial run and using a single balancing plane. The method thus avoids multiple trial runs required for modal balancing of flexible rotors. This method is verified on an experimental rotor having both bow and unbalance. The concept of quantifying the distributed unbalance using ‘Norm’ of eccentricity polynomial function is also introduced for the first time.