A recurrent stochastic binary network

Stochastic neural networks are usually built by introducing random fluctuations into the network. A natural method is to use stochastic connections rather than stochastic activation functions. We propose a new model in which each neuron has very simple functionality but all the connections are stochastic. It is shown that the stationary distribution of the network uniquely exists and it is approxi-mately a Boltzmann-Gibbs distribution. The relationship between the model and the Markov random field is discussed. New techniques to implement simulated annealing and Boltzmann learning are pro-posed. Simulation results on the graph bisection problem and image recognition show that the network is powerful enough to solve real world problems.     

Analysis of surveillance test interval by Markov process for SDS1 in CANDU nuclear power plants

The Canadian Nuclear Safety Commission (CNSC) requires that each shutdown system (SDS) of CANDU plant should be available more than 99.9% of the reactor operating time and be tested periodically. The compliance with the availability requirement should be demonstrated using the component failure rate data and the benefits of the tests. There are many factors that should be considered in determining the surveillance test interval (STI) for the SDSs. These includes: the desired target availability, the actual unavailability, the probability of spurious trips, the test duration, and the side effects such as wear-out, human errors, and economic burdens. A Markov process model is developed to study the effect of test interval in the shutdown system number one (SDS1) in this paper. The model can provide the quantitative data required for selecting the STI. Representing the state transitions in the SDS1 by a time-homogeneous Markov process, the model can be used to quantify the effect of surveillance test durations and interval on the unavailability and the spurious trip probability. The model can also be used to analyze the variation of the core damage probability with respect to changes in the test interval once combined with the conditional core damage model derived from the event trees and the fault trees of probabilistic safety assessment (PSA) of the nuclear power plant (NPP).     

Bicharacteristic quality control in manufacturing

In this paper, the importance of bicharacteristic quality control has been discussed, when it is desired to exercise process control on manufacturing products that present two measurable characteristics which can suffer variations and on the other hand are considered significant for the functioning or the acceptance of the product. The verification that the process is in control or not is made by examining periodically the items of the samples drawn and plotting the points representing the mean of the characteristic in analysis on the corresponding charts. Using the computational program developed, a practical application is carried out for a mechanical manufacturing component that presents two critical characteristics. Also, a programmable mathematical model has been developed, aiming to computationally determine the sample size, and the interval between successive samples, which minimize the expected total cost. Idealized numerical application illustrates the use of the cost model that optimizes these parameters.     

A stochastic control model for optimal timing of climate policies

A stochastic control model is proposed as a paradigm for the design of optimal timing of greenhouse gas (GHG) emission abatement. The resolution of uncertainty concerning climate sensitivity and the technological breakthrough providing access to a carbon-free production economy are modeled as controlled stochastic jump processes. The optimal policy is characterized using the dynamic programming solution to a piecewise deterministic optimal control problem. A numerical illustration is developed with a set of parameters calibrated on recently proposed models for integrated assessment of climate policies. The results are interpreted and the insights they provide on the timing issue of climate policy are discussed.     

Attention switching during scene perception: How goals influence the time course of eye movements across advertisements.

Eye movements across advertisements express a temporal pattern of bursts of respectively relatively short and long saccades, and this pattern is systematically influenced by activated scene perception goals. This was revealed by a continuous-time hidden Markov model applied to eye movements of 220 participants exposed to 17 ads under a free-viewing condition, and a scene-learning goal (ad memorization), a scene-evaluation goal (ad appreciation), a target-learning goal (product learning), or a target-evaluation goal (product evaluation). The model reflects how attention switches between two states--local and global--expressed in saccades of shorter and longer amplitude on a spatial grid with 48 cells overlaid on the ads. During the 5- to 6-s duration of self-controlled exposure to ads in the magazine context, attention predominantly started in the local state and ended in the global state, and rapidly switched about 5 times between states. The duration of the local attention state was much longer than the duration of the global state. Goals affected the frequency of switching between attention states and the duration of the local, but not of the global, state. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Analysis of long flow lines with quality and operational failures

This paper presents approximation methods for the performance analysis of long manufacturing lines, i.e. lines with more than two machines and one buffer, that have both quality and operational failures. We describe three different versions of long flow lines that differ in the locations of the inspection stations and in the sets of machines that each inspection station monitors. We explain a transformation method that approximates long manufacturing lines that have quality and operational failures with long lines that only have operational failures. Such lines can be evaluated by decomposition methods. We introduce other approximations to quantify the effects of the separation of inspections from operations. Comparison with simulation shows that the solution methods provide reliable performance estimates.     

Modelling and estimating the reliability of stochastic dynamical systems with Markovian switching

In this paper, a general framework for the modelling of physical phenomena with stochastic dynamical systems switched by jump Markov processes is given. A methodology of the associated estimation procedures is provided. A particular attention is paid to the estimation of the underlying jump process, which is not observable.As an application, a stochastic model is proposed for the fatigue crack growth problem. The estimation of the model parameters is made on a real crack growth data set. We are thus able to simulate some crack growth paths which are used for reliability analysis through Monte Carlo techniques.     

Scalable Schemes for Call Admission and Handover in Cellular Networks with Multiple Services

The introduction of new service categories withdifferent bandwidth requirements, e.g., data and multimedia, to cellularmobile radio networks makes many of the traditional mechanisms for controlingtraffic unusable orless efficient. The call admission and the handover handling are of the mostsensitive issues in this extension to new services. The performance of allservices includingthe traditional voice and the new services can be dramatically affected ifappropriate schemes are not used. In this paper, we propose call admission andhandover handling schemes for a cellular mobile network that offers twoservice types: voice and data. The data connections are assumed to transmitatdifferent transmission rates that are integer multiples to that of one radiochannel. In the case of congestion, the base station asks the active dataconnections to reduce their transmission rate in order to provide freechannels for the newly arrived request of both service types. This isbasically intended for incoming handover requests. The request will berejected if the transmission rate of the active connections reaches a givenminimum rate. Similar mechanism can also be used for new call arrivals, butsome priority can be given to handovers by setting a higher transmission ratethreshold for the new call rejection. As an extension to the proposedscalability, aqueuing of new calls is also proposed and analyzed. Analytical models werebuilt for the two proposed schemes together with the traditional channelreservation scheme. The effect of different traffic and configurationparameters on the performance measures like the grade of service, blockingprobabilities, and utilization, are studied using the proposed technique.Results show that the proposed schemes provide very good performance and morefairness among the different service types.     

Performance modeling of a manufacturing cell with a single material transporter: An approximation

We consider a flexible manufacturing system with a number of workstations, a single material transporter, and a common storage space of finite capacity. The material handling delay times are explicitly considered in the model and assumed to follow a two-stage Coxian distribution. The material processing times on a workstation also have a two-stage Coxian distribution. The routing of parts within the system follows a Markov chain. An approximate performance model is developed and the results are compared with the exact or simulation results. We also investigate how this performance model compares to a simulation with deterministic routing and processing times. Finally, we study the effect, on the performance measures, of ignoring the material transporter or of modeling the transporter as a central server with aggregation of routing information.     

对数马尔可夫模型在乌鲁木齐市降水预测中的应用

中长期降水量的预测是气象科学的一个难点问题,也是水文学中的一个重要问题。建立对数马尔可夫模型预测降水量,弥补了传统的马尔可夫模型降水预测中峰值的不准确性,提高了预测精度,并用乌鲁木齐市气象站43年降水资料进行了验证。结果表明,模型预测精度较高,为干旱半干旱区中长期降水量预报提供了一条简便可行的途径。