ATM network design and optimization at the call-level may be formulated in the framework of multirate, circuit-switched, loss networks with effective bandwidth encapsulating cell-level behavior. Each service supported on the ATM network is characterized by a rate or bandwidth requirement. Future networks will be characterized by links with very large capacities in circuits and by many rates. Various asymptotic results are given to reduce the attendant complexity of numerical calculations. A central element is a uniform asymptotic approximation (UAA) for link analyses. Moreover, a unified hybrid approach is given which allows asymptotic and nonasymptotic methods of calculations to be used cooperatively. Network loss probabilities are obtained by solving fixed point equations. A canonical problem of route and logical network design is considered. An optimization procedure is proposed, which is guided by gradients obtained by solving a system of equations for implied costs. A novel applic...

Consider a \black box " having I input channels and J output channels. Each arrival on an input channel gets routed through the black box and appears on an output channel. The system is monitored for a xed time period and a record is made of the number of arrivals on each input channel and the number of departures on each output channel. The OD (origination-destination) matrix estimation problem is to estimate, for each i and j, the number of arrivals on channel i that depart on channel j. Weintroduce a Poisson stochastic model and employ the EM algorithm to produce high likelihood estimates. In the case of estimation based on observations over a single time-period, we analyze in detail the xed points of the EM algorithm showing that every vertex of a certain polytope of feasible matrices is a xed point and identifying a speci c interior xed point which is a saddle point for the likelihood function.

A hybrid model based mostly on a high-order Markovchain and occasionally on an independence model is proposed for pro#ling the command-sequence of a computer user in order to identify a "signature behavior" for that user. Based on the model, an estimation procedure for such a signature behavior driven by Maximum Likelihood #ML# considerations is devised. The formal ML estimates are numerically intractable, but the ML-optimization problem can be substituted by a linear inverse problem with positivity constraints #LININPOS#, for which the EM algorithm can be used as an equation solver to produce an approximate ML-estimate. A user's command-sequence is then compared to his and others' estimated signature-behavior in real time, by means of statistical hypothesis testing. A form of the likelihood-ratio test is used to test if a given sequence of commands is from the proclaimed user, with the alternative hypothesis being masquerader user. Data from a real-life experiment, conducted at a research lab, is used to assess the method. Key Words: Anomaly Detection; Unix; Mixture Transition Distribution #MTD#; LININPOS; EM. 1

The main object of this paper is to present some general concepts of Bayesian inference and more specifically the estimation of the hyperparameters in inverse problems. We consider a general linear situation where we are given some data y related to the unknown parameters a by y = Aa + n and where we can assign the probability laws p(a10), p(yl, ), p() and p(O). The main discussion is then how to infer a, 0 and fl either individually or any combinations of them. Different situations are considered and discussed. As an important example, we consider the case where 0 and fl are the precision parameters of the Gaussian laws to whom we assign Gamma priors and we propose some new and practical algorithms to estimate them simultaneously. Comparisons and links with other classical methods such as maximum likelihood are presented.

Abstract not found

Abstract. With increasingly more databases becoming available on the Internet, there is a growing opportunity to globalise knowledge discovery and learn general patterns, rather than restricting learning to specific databases from which the rules may not be generalisable. Clustering of distributed databases facilitates learning of new concepts that characterise common features of, and differences between, datasets. We are here concerned with clustering databases that hold aggregate count data on a set of attributes that have been classified according to heterogeneous classification schemes. Such aggregates are commonly used for summarising very large databases such as those encountered in data warehousing, large-scale transaction management, and statistical databases. For measuring difference between aggregates we utilise two distance metrics: the Euclidean distance and the Kullback-Leibler information divergence. A hybrid between Kullback-Leibler and the Euclidean distance, which uses the former to learn the class probabilities and the latter as the corresponding distance measure, looks particularly promising both in terms of accuracy and scalability. These metrics are evaluated using synthetic data. Important applications of the work include the clustering of heterogeneous customer databases for the discovery of new marketing concepts and the clustering of medical databases for the discovery of new epidemiological concepts. 1.

Abstract — Like other problems in network tomography or traffic matrix estimation, the location of congested IP links from end-to-end measurements requires solving a system of equations that relate the measurement outcomes with the variables representing the status of the IP links. In most networks, this system of equations does not have a unique solution. To overcome this critical problem, current methods use the unrealistic assumption that all IP links have the same prior probability of being congested. We find that this assumption is not needed, because these probabilities can be uniquely identified from a small set of measurements by using properties of Boolean algebra. We can then use the learnt probabilities as priors to find rapidly the congested links at any time, with an order of magnitude gain in accuracy over existing algorithms. We validate our results both by simulation and real implementation in the PlanetLab network. I.

The design and analysis of an algorithm for the restoration of degraded images of machine-printed characters is presented. The input is a set of degraded bilevel images of a single unknown character; the output is an approximation to the character 's ideal artwork. The algorithm seeks to minimize the discrepancy between the approximation and the ideal, measured as the worst-case Euclidean distance between their boundaries. We investigate a family of algorithms which superimpose the input images, add up the intensities at each point, and threshold the result. We show that, under degradations due to random spatial sampling error, significant asymptotic improvements can be achieved by suitably preprocessing each input image and postprocessing the final result. Experimental trials on special test shapes and Latin characters are discussed. 1 Introduction In the last few years, a variety of document-image degradation models have been proposed, and their applications investigated [3]. Models...

This paper describes a parametric deconvolution method(PDPS) appropriate for a particular class of signals which we call spike-convolution models. These models arise when a sparse spike train---Dirac deltas according to our mathematical treatment---is convolved with a fixed point-spread function, and additive noise or measurement error is superimposed. We view deconvolution as an estimation problem, regarding the locations and heights of the underlying spikes, as well as the baseline and the measurement error variance as unknown parameters. Our estimation scheme consists of two parts: model fitting and model selection. To fit a spikeconvolution model of a specific order, we estimate peak locations by trigonometric moments, and heights and the baseline by least squares. The model selection procedure has two stages. Its first stage is so designed that we expect a model of a somewhat larger order than the truth to be selected. In the second stage, the final model is obtained using backwar...

The problem of variance assessment for Positron Emission Tomography (PET) image reconstructions is considered in the context of generalized deconvolution. A refinement of an approximate technique proposed by Carson et al. (1993) is examined. Computational implications of representing the reconstruction kernel in terms of a weighted sum of Gaussian densities are developed. Bias and variance characteristics of the resulting variance estimators are examined by numerical simulation. For typical regions, the error in estimated standard deviations is found to be on the order of 10%. The use of smoothing to obtain more reliable point-wise variance estimators is described and some theoretical analysis of this technique is carried out. For the PET application, simulations suggest that the percent improvement in the root mean squared error accuracy of point-wise variance estimators obtained by smoothing can be on the order of 30%. A practical application of the methodology to a PET study is pres...