Pasquale Sullo

Estimating the Parameters of a Multivariate Exponential Distribution

Abstract Parameter estimation for a (k + 1)-parameter version of the k-dimensional multivariate exponential distribution (MVE) of Marshall and Olkin is investigated. Although not absolutely continuous with respect to Lebesgue measure, a density with respect to a dominating measure is specified, enabling derivation of a likelihood function and likelihood equations. In general, the likelihood equations, not solvable explicitly, have a unique root which is the maximum likelihood estimator (MLE). A simple estimator (INT) is derived as the first iterate in solving the likelihood equations iteratively. The resulting sequence of estimators converges to the MLE for sufficiently large samples. These results can be extended to the more general (2 k − 1)-parameter MVE.

Poisson mixture yield models for integrated circuits: A critical review

Yield models for semiconductor devices are often derived from Poisson mixtures, the specific model being dependent on the distribution assumed for the density of point defects. Presented here is a review of the development of such yield models, including two new models obtained by using the Rayleigh and the inverse Gaussian as mixing distributions. The various models are compared and some general properties of yield functions for any Poisson mixture are derived. The concept of reference region is introduced and its implication for yield modeling is discussed.

Reliability of communication flow in R&D organizations

Much attention has been given to the communication process in R&D organizations since the concept of a technological gatekeeper was proposed. By viewing the R&D organization as an information processing and generating system, the role communication network structure plays in determining R&D performance can be isolated and studied. Findings from empirical research are reviewed, providing a framework in which to examine and assess the communication patterns present in an R&D organization. A methodology is presented to evaluate the effectiveness of an organizational communication network with particular reference to project management. The proposed method permits the assessment of contemplated management actions intended to improve organizational communication.

A Reliability Theoretic Construct for Assessing Information Flow in Networks.

A reliability theoretic construct is proposed for conceptualizing the process of information flow. It focuses on information produced to satisfy specified purposes or to achieve preconceived objectives. Furthermore, the model incorporates explicitly the concept of an information producer contemplating a choice of action in an uncertain environment. The resulting models are therefore prescriptive in nature. The usefulness of this construct is illustrated by a case analysis of the effectiveness of natural resource data products in land-use decision making. Measures of system reliability of the information flow network are determined and sensitivity analyses performed. Numerical examples are presented and discussed. The prescriptive nature of this approach permits use of its results to indicate how a data producer can increase the effectiveness of documents by identifying the information flow network, assessing the reliability of each component in the network, finding measures of system reliability, and performing sensitivity analyses to identify the critical components of the system. The result is a closer congruence between the objectives of the data producer and the requirements of users.

Estimating performance degradation using intervals between upcrossings of a threshold

We consider the problem of estimating the rate of degradation of a system when only limited data is available. Specifically, we consider the problem of deriving a consistent estimator of degradation when we can only observe the sequence of times when an unobserved performance measurement exceeds a given threshold. Such problems arise, for example, in the operation of complex systems where exceeding a predefined level multiple times drives repair actions, in security applications for which only entry into a secured facility is recorded, and in sensor networks where data acquisition and transmission are limited. We consider the relative efficiency of the resulting estimator when compared to the MLE estimators derived from the full unobserved continuous process and from unobserved binary high-low data.