自由空间光学Comega多级互连网络特性研究

光学Ｃｏｍｅｇａ网络是一种新颖的易于光学实现的多级光互连网络。本文对该网络结构及特性进行了详细分析，和到了网络的互连函数，并给出了它与Ｓｔａｒａｎ，Ｃｒｏｓｓｏｖｅｒ，Ｏｍｅｇａ，Ｂａｎｙａｎ及基准等常见网络的拓扑等价证明。     

数字光开关网络中四功能交换开关光学Omega网络

本文提出了一种用于数字光开关网络中，四功能交换开关光学Ｏｍｅｇａ网络。文中详细地阐述了用液晶空间光调制方法实现光学四功能交换开关，和Ｏｍｅｇａ互连网络的原理，并给出了光学系统和实验结果。     

榕树网络的互连矩阵与模拟研究

本文导出了榕树网络的互连矩阵，并进行了模拟计算，结果与实验吻合。从而为理论上研究榕树网络的互连函数提供了一种有效的分析方法。     

光互连Omega网络与全交叉网络拓扑等价的多样性

本文对互连网络拓朴等价的分析方法作了简单评述,并采用互连网络拓朴等价的图分析法分析了自由空间光学互连Omega网络与全交叉网络的拓朴等价及其多样性,拓展了Omega网络在通信交换、数字光计算以及并行多处理机系统的潜在应用。     

榕树型互连网络及光学实现

讨论了一种用扭曲向列相液晶器件和偏振分光棱镜实现光学榕树上互连网络的结构，这种结构具有直线式传输、无错位、无缩放，并能够实现模块化和集成化的优点。     

未来计算系统中的光互连

论述了光互连网络对巨型计算机和智能计算机发展的重要作用，指出了光互连网络是解决目前电子计算机中通信“瓶颈”的关键，讨论了采用光互连的物理依据，强调了光互连网络的高效，快速和大容量特性。最后分析了我们正在研究的光学交叉开关互连网络的几个关键参数和初步实验结果。     

反-逆混洗光电混合循环排序网

提出了一种新型光电混合循环排序网络。排序网的互连级采用自由空间光学反－逆混洗（Ｃｏｍｅｇａ）互连,比较交换节点列阵采用硅互补金属－氧化物－半导体－自由光效应器件（ＣＭＯＳ－ＳＥＥＤ）光电混合集成电路来实现。由于反－逆混洗多级网络各互连级完全相同,该排序网络通过采用循环结构大大地简化系统所需的硬件。通过采用并行比较交换节点进一步提高了排序网络中比较操作的速度。     

并行处理阵列中光电混合光纤Crossbar互连网络的研制

本文综述了用于三维光电混合并行处理器阵列系统中的双向可编程多样重构光纤Crossbar互连网络。重点介绍了基于光纤通信的Crossbar互连网络中光纤互连器的设计原理与制作方法,并给出一种具体的电路图,用自制的伪随机码发生器测试了器件的整体性能,而后进行了应用性联机测试。结果表明,该光纤互连器能很好地满足系统中处理器20Mbps传输速率的要求,电路简单,具有通用性和较高的性能价格比。     

自由空间光互连的发展

介绍自由空间光互连技术的发展。评述如何利用自由空间光互连技术构成互连网络和利用自由空间光互连技术在二维光电逻辑元件中实现三维光互连的早期研究、目前自由空间光互连技术在VLSI中的应用以及自由空间光互连的前景。     

全交叉互连的模拟实现

采用互连矩阵对全交叉网络的互连函数进行了模拟计算,计算结果与实验吻合。从而为验证光学全交叉网络的实验结果以及模拟实现全交叉网络的互连函数提供了理论依据。     

Blocking probability analysis of pruned optical banyan networks on vertical stacking scheme with extra banyan planes

Vertically stacked pruned optical banyan networks with extra planes (in short, EP-VSOB networks) have lower switch count and optimal time complexity (O(log₂ N)) for routing N input requests. However, blocking probability is relatively higher than that of a VSOB networks using regular banyan planes. In the EP-VSOB architecture, the number of pruned planes has always been considered as , and a few extra planes (regular banyan) have been added with these pruned planes. In this paper we present the results of blocking analysis of a more generalized architecture in which the number of pruned planes can be 2^x, where x ? 0 in addition to the variable extra planes. This generalization helps us make a compromise between different constraints and performance metrics. Our simulation results show that for some given performance requirements (e.g. cost, speed or blocking probability) we can choose a network that has lower switch count compared to -plane pruned crosstalk-free optical banyan networks. For example, to ensure blocking probability <0.02, previously we would chose a pruned network of 32 pruned planes and 1 extra planes (a regular banyan); however, our simulations results show that a network of 16 pruned planes and 2 extra planes is enough to ensure the same performance. It is notable that, the hardware cost decreases by 28.65% in this new combination of pruned and extra planes. We believe our results will provide more flexibility in choosing a particular EP-VSOB network satisfying given requirements.     

Synchronization between Different Networks

Synchronization between two networks with different topology structures and different dynamical behaviours is studied. These two different networks are driving and responding networks, respectively. Under the preconditions that the driving network gets synchronization, we give the conditions for the responding network to be synchronized to the same dynamics as the driving network with the help of the open-plus-closed-loop method. Then a example is given to verify the validity of the theoretical results.     

Pruned optical banyan networks on vertical stacking scheme for faster connection establishment

In this paper, we address the issue of faster connection establishment in a large vertically stacked optical Banyan (VSOB) network. The best known global routing algorithm, which turns an N × N crosstalk-free VSOB network into a rearrangeably non-blocking one, has time complexity O (NlogN). This is quite large compared to O (NlogN) time complexity of a single plane banyan network, which is a self-routing network with very high blocking probability. For a large size of switching network this O (NlogN) time complexity may result unacceptably long delay. Therefore, an optical network with very low blocking probability and O (NlogN) time complexity will be useful. Previously proposed Plane Fixed Routing (PFR) algorithm has O (logN) time complexity but results in higher than 2% blocking probability with zero-crosstalk constraint for a network as large as 4096 × 4096 at full load. In this paper, first we propose the pruning of VSOB networks that reduces the hardware cost by almost 30%. The networks can still use the PFR algorithm and results in the same blocking probability. However, we show that the blocking probability can be reduced dramatically while keeping the optimum time complexity O (logN) by allowing only a small amount of crosstalk. Then, we propose a new kind of switching networks in which extra regular banyan planes have been added with the pruned VBOS (P-VSOB) networks. Necessary routing algorithms, namely, PFR_RS and PFR_LS show that this new switching network can reduce the blocking probability to very low value even with zero-crosstalk constraint while keeping the hardware cost 3almost the same as for P-VSOB networks. Both these algorithms also have time complexity O (NlogN).     

Neural Networks Retrieving Boolean Patterns in a Sea of Gaussian Ones

Restricted Boltzmann machines are key tools in machine learning and are described by the energy function of bipartite spin-glasses. From a statistical mechanical perspective, they share the same Gibbs measure of Hopfield networks for associative memory. In this equivalence, weights in the former play as patterns in the latter. As Boltzmann machines usually require real weights to be trained with gradient-descent-like methods, while Hopfield networks typically store binary patterns to be able to retrieve, the investigation of a mixed Hebbian network, equipped with both real (e.g., Gaussian) and discrete (e.g., Boolean) patterns naturally arises. We prove that, in the challenging regime of a high storage of real patterns, where retrieval is forbidden, an additional load of Boolean patterns can still be retrieved, as long as the ratio between the overall load and the network size does not exceed a critical threshold, that turns out to be the same of the standard Amit–Gutfreund–Sompolinsky theory. Assuming replica symmetry, we study the case of a low load of Boolean patterns combining the stochastic stability and Hamilton-Jacobi interpolating techniques. The result can be extended to the high load by a non rigorous but standard replica computation argument.     

Detecting Communities by Revised Max-flow Method in Networks

A ubiquitous phenomenon in networks is the presence of communities within which the network connections are dense and between which they are sparser.This paper proposes a max-flow algorithm in bipartite networks to detect communities in general networks.Firstly,we construct a bipartite network in accordance with a general network and derive a revised max-flow problem in order to uncover the community structure.Then we present a local heuristic algorithm to find the optimal solution of the revised max-flow problem.This method is applied to a variety of real-world and artificial complex networks,and the partition results confirm its effectiveness and accuracy.     

Detecting Communities by Revised Max-flow Method in Networks

A ubiquitous phenomenon in networks is the presence of communities within which the network connections are dense and between which they are sparser. This paper proposes a max-flow algorithm in bipartite networks to detect communities in general networks. Firstly, we construct a bipartite network in accordance with a general network and derive a revised max-flow problem in order to uncover the community structure. Then we present a local heuristic algorithm to find the optimal solution of the revised max-flow problem. This method is applied to a variety of real-world and artificial complex networks, and the partition results confirm its effectiveness and accuracy.     

增长及非增长无标度网络的成因解析

分析两类无标度网络的形成原因,提出一个无标度网络演化模型并进行一系列数值实验.基于分析和实验得到推论:只要保持足够低的网络密度,通过基于度的偏好连接就可形成长期稳定的无标度网络.规模增长和点边增删既是客观存在,又起到了控制网络密度的作用,足够低的网络密度和基于度的偏好连接是所有无标度网络共同的必要条件.推论可同时解释增长和非增长无标度网络的形成原因.研究结果有助于理解各种真实无标度网络和建立相应的模型.     

Nonlocal Mechanism for Synchronization of Time Delay Networks

We present the interplay between synchronization of networks with heterogeneous delays and the greatest common divisor (GCD) of loops composing the network. We distinguish between two types of networks; (I) chaotic networks and (II) population dynamic networks with periodic activity driven by external stimuli. For type (I), in the weak chaos region, the units of a chaotic network characterized by GCD=1 are in a chaotic zero-lag synchronization, whereas for GCD>1, the network splits into GCD-clusters in which clustered units are in zero-lag synchronization. These results are supported by simulations of chaotic systems, self-consistent and mixing arguments, as well as analytical solutions of Bernoulli maps. Type (II) is exemplified by simulations of Hodgkin Huxley population dynamic networks with unidirectional connectivity, synaptic noise and distribution of delays within neurons belonging to a node and between connecting nodes. For a stimulus to one node, the network splits into GCD-clusters in which cluster neurons are in zero-lag synchronization. For complex external stimuli, the network splits into clusters equal to the greatest common divisor of loops composing the network (spatial) and the periodicity of the external stimuli (temporal). The results suggest that neural information processing may take place in the transient to synchronization and imply a much shorter time scale for the inference of a perceptual entity.