Hopfield-Like Neural Nets and Sensor Networks

The transformation of a sensor network (SN) into a neural Hopfield-like network (HLN) is proposed. The SN of interest is a nonlinear non-reciprocal population of coupled oscillators. The proposed transformation is useful for investigating the relation between the structure of the SN and its capability of arriving to a global consensus. The case of a 3-SN is developed in detail for illustrating the advantages of the suggested transformation. Both the structural conditions necessary for achieving in this case the consensus and its relation to local measurements are presented.     

模糊Petri网的反馈控制研究

本文首先将模拟技术应用于Ｐｅｔｒｉ网中，得到模糊Ｐｅｔｒｉ网模型，然后基于Ｐｅｔｒｉ网中的库所湾量的概念，在普通Ｐｅｔｒｉ网的反馈控制基础上提出了一种模型Ｐｅｔｒｉ网的反馈控制方法。该方法使得对复杂系统的模糊Ｐｅｔｒｉ多控制器的系统设计成为可能。     

Discovering Text Databases with Neural Nets

Since documents on the Web are naturally partitioned into many text databases, the efficient document retrieval process requires identifying the text databases that are most likely to provide relevant documents to the query and then searching for the identified text databases. In this paper, we propose a neural net based approach to such an efficient document retrieval. First, we present a neural net agent that learns about underlying text databases from the user's relevance feedback. For a given query, the neural net agent, which is sufficiently trained on the basis of the BPN learning mechanism, discovers the text databases associated with the relevant documents and retrieves those documents effectively. In order to scale our approach with the large number of text databases, we also propose the hierarchical organization of neural net agents which reduces the total training cost at the acceptable level. Finally, we evaluate the performance of our approach by comparing it to those of the conventional well-known approaches. Received 5 March 1999 / Revised 7 March 2000 / Accepted in revised form 2 November 2000     

Feedback Control Logic for Backward Conflict Free Choice Nets

This paper discusses the forbidden state problem, as specified by generalized mutual exclusion constraints, in the context of supervisory control of discrete event systems modelled by Petri nets. The case of backward-conflict-free and free-choice uncontrollable subnets is considered and it is shown how to transform such subnets in well-formed free-choice nets. Then, the well-formed free-choice nets are decomposed in marked graph components by recurring to minimal T-invariants. The forbidden state problem is so reformulated for the obtained marked graph components into an equivalent one which is shown to be a linear programming problem. Thus, improving existing results in literature, a polynomial complexity solution, suitable for on-line control, is achieved. Free-choice relationship and cycle modelling, that frequently occur in real-life situations, are so allowed in the uncontrollable subnet     

Complexity Results on Learning by Neural Nets

We consider the computational complexity of learning by neural nets. We are interested in how hard it is to design appropriate neural net architectures and to train neural nets for general and specialized learning tasks. Our main result shows that the training problem for 2-cascade neural nets (which have only two non-input nodes, one of which is hidden) is N-complete, which implies that finding an optimal net (in terms of the number of non-input units) that is consistent with a set of examples is also N-complete. This result also demonstrates a surprising gap between the computational complexities of one-node (perceptron) and two-node neural net training problems, since the perceptron training problem can be solved in polynomial time by linear programming techniques. We conjecture that training a k-cascade neural net, which is a classical threshold network training problem, is also N-complete, for each fixed k2. We also show that the problem of finding an optimal perceptron (in terms of the number of non-zero weights) consistent with a set of training examples is N-hard.Our neural net learning model encapsulates the idea of modular neural nets, which is a popular approach to overcoming the scaling problem in training neural nets. We investigate how much easier the training problem becomes if the class of concepts to be learned is known a priori and the net architecture is allowed to be sufficiently non-optimal. Finally, we classify several neural net optimization problems within the polynomial-time hierarchy.     

神经网络在话务异常分析中的应用

在论述电信话务异常分析的任务和难点的基础上，试图将神经网络信息处理方法引入话务异常分析中，给出了话务异常分析的一般神经网络模型结构，并给出了一些实际测试结果。     

Neural Nets: An Evaluation and a Spreadsheet Implementation

Attitudes to neural nets range from suspicion to uncritical admiration. This paper aims to introduce nets and to evaluate their strengths and weaknesses. The language is non-technical, but the conceptual treatment is intended to be rigorous. A practical method for implementing a neural net on a spreadsheet is described, and sample results illustrated.     

Incremental Evolution in ANNs: Neural Nets which Grow

This paper explains the optimisation of neuralnetwork topology using Incremental Evolution;that is, by allowing the network to expand byadding to its structure. This method allows anetwork to grow from a simple to a complexstructure until it is capable of fulfilling itsintended function. The approach is somewhatanalogous to the growth of an embryo or theevolution of a fossil line through time, it istherefore sometimes referred to as anembryology or embryological algorithm. Thepaper begins with a general introduction,comparing this method to other competingtechniques such as The Genetic Algorithm, otherEvolutionary Algorithms and SimulatedAnnealing. A literature survey of previous workis included, followed by an extensive newframework for application of the technique.Finally, examples of applications and a generaldiscussion are presented.     

Information and Meaning: Use-Based Models in Arrays of Neural Nets

The goal of philosophy of information is to understand what information is, how it operates, and how to put it to work. But unlike information in the technical sense of information theory, what we are interested in is meaningful information. To understand the nature and dynamics of information in this sense we have to understand meaning. What we offer here are simple computational models that show emergence of meaning and information transfer in randomized arrays of neural nets. These we take to be formal instantiations of a tradition of theories of meaning as use. What they offer, we propose, is a glimpse into the origin and dynamics of at least simple forms of meaning and information transfer as properties inherent in behavioral coordination across a community.     

Refractoriness in Poisson and Gaussian First-order Neural Nets with Chemical Markers

In this work first order probabilistic Poisson and Gaussian neural nets with chemical markers are investigated, analytically and by computer simulations. The investigation of steady-state behavior of these systems is extended here to systems in which the refractory period is assigned to be 1 for all or some of the subpopulations of the net, whereas the remainder are characterized by zero refractory periods. The interest is focused on the effects of refractoriness on the neural activities. Results obtained show the existence of several critical points at high initial activities, which are a consequence of the nonzero refractory periods. For these points a larger initial activity, above a certain critical level, results in the reduction of activity to a lower stable steady-state, instead of the highest one. We also find that in the Gaussian nets each critical point is lower than the corresponding one as in the Poisson nets. Finally, a discussion of the results is made.     

网简化技术在Petri网反馈控制器设计中的应用

针对基于Petri网离散事件系统库所标识不等式约束反馈控制器的设计问题,提出一项新型网简化技术--标识总量保持网简化技术,并利用此项技术把所有约束库所融合为一个与它们标识总量相等的库所,使控制器的设计更为简单,尤其是对规模较大的系统,其优势更加明显.借助一个应用实例,将该控制器设计方法与Moody等人提出的控制器设计方法作比较,表明所提出的控制器设计方法更简单、更有效.     

Neural Nets Trained by Genetic Algorithms for Collision Avoidance

As air traffic keeps increasing, many research programs focus on collision avoidance techniques. For short or medium term avoidance, new headings have to be computed almost on the spot, and feed forward neural nets are susceptible to find solutions in a much shorter amount of time than classical avoidance algorithms (A ^*, stochastic optimization, etc.) In this article, we show that a neural network can be built with unsupervised learning to compute nearly optimal trajectories to solve two aircraft conflicts with the highest reliability, while computing headings in a few milliseconds.     

一种神经网络体视协同算法

描述一种多分辨率体视协同算法.该算法用二分正交小波作多分辨率分解,采取边匹配边重建,二者互相修正协同进行的策略,并以神经网络主要是Hopfield网络为其实现形式.实验结果表明.该算法是合理和有效的.     

一种用于计算机模拟的神经元模型

本文提出了一种单个神经元的模型,它能较好地解决计算机模拟语言所要求的通用性和精确性之间的关系问题,通过引入S型曲线来描述突触上的记忆值,能给短时记忆和长时记忆一个统一的刻画,并在突触水平上模拟了习惯化、敏感化、条件化和容易化等四种基本学习形式.     

一种神经网络判决反馈均衡器

在分析Chebyshev正交多项式神经网络非线性滤波器的基础上,利用判决反馈均衡器的结构特点,提出了一种Chebyshev正交多项式神经网络判决反馈均衡器,给出了对应的自适应NLMS算法.数值仿真结果表明,该均衡器结构能够更有效地消除码间和非线性干扰,降低信号的误码率.     

一类受控Petri网的状态反馈逻辑的综合

将Holloway和krogh关于受控标记图的禁态控制方面的结果扩展到更广泛的一类受控Petri网--不可控子网为前后向无冲突的受控Petri网,并去掉了关于初始标记和禁态规范的限制.     

Probabilistic and Other Neural Nets in Multi-Hole Probe Calibration and Flow Angularity Pattern Recognition

The use of probabilistic (PNN) and multilayer feedforward (MLFNN) neural networks is investigated for the calibration of multi-hole pressure probes and the prediction of associated flow angularity patterns in test flow fields. Both types of network are studied in detail for their calibration and prediction characteristics. The current formalism can be applied to any multi-hole probe, however the test results for the most commonly used five-hole Cone and Prism probe types alone are reported in this paper. Received: 1 October 1998?Received in revised form: 12 December 1998?Accepted: 16 December 1998     

广义连续反馈型神经网络的稳定性

通过对连续Hopfield网络模型的抽象,提出了广义连续反馈型神经网络的概念,指出一种不需要构造能量函数对网络模型的稳定性进行分析和判别的方法,同时给出了相应的数学证明。     

Adaptive Feedback Linearization Using Efficient Neural Networks

For a class of single-input, single-output, continuous-time nonlinear systems, a feedback linearizing neural network (NN) controller is presented. Control action is used to achieve tracking performance. The controller is composed of a robustifying term and two neural networks adapted on-line to linearize the system by approximating two nonlinear functions. A stability proof is given in the sense of Lyapunov. No off-line weight learning phase is needed and initialization of the network weights is straightforward. The NN controller is tested on a standard benchmark problem.