• Journal of Information Processing Systems
• /
• v.8 no.2
• /
• pp.191-212
• /
• 2012

Since a social network by definition is so diverse, the problem of estimating the preferences of its users is becoming increasingly essential for personalized applications, which range from service recommender systems to the targeted advertising of services. However, unlike traditional estimation problems where the underlying target distribution is stationary; estimating a user's interests typically involves non-stationary distributions. The consequent time varying nature of the distribution to be tracked imposes stringent constraints on the "unlearning" capabilities of the estimator used. Therefore, resorting to strong estimators that converge with a probability of 1 is inefficient since they rely on the assumption that the distribution of the user's preferences is stationary. In this vein, we propose to use a family of stochastic-learning based Weak estimators for learning and tracking a user's time varying interests. Experimental results demonstrate that our proposed paradigm outperforms some of the traditional legacy approaches that represent the state-of-the-art technology.

• Structural Engineering and Mechanics
• /
• v.49 no.3
• /
• pp.411-425
• /
• 2014

This paper presents the processing of nonlinear features associated with a damage event by quadratic time-frequency distributions for damage identification in a frame structure. A time-frequency distribution is a function which distributes the total energy of a signal at a particular time and frequency point. As the occurrence of damage often gives rise to non-stationary, nonlinear structural behavior, simultaneous representation of the dynamic response in the time-frequency plane offers valuable insight for damage detection. The applicability of the bilinear time-frequency distributions of the Cohen class is examined for the damage assessment of a frame structure from the simulated acceleration data. It is shown that the changes in instantaneous energy of the dynamic response could be a good damage indicator. Presence and location of damage can be identified using Choi-Williams distribution when damping is ignored. However, in the presence of damping the Page distribution is more effective and offers better readability for structural damage detection.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.7 no.2
• /
• pp.308-326
• /
• 2013

Mobility Model has drawn more and more attentions since its critical role in Mobile Wireless Networks performance evaluation. This paper analyzes the mobility patterns in the catastrophe rescue scenario, and proposes the Random Waypoint with Base Point mobility model to model these characteristics. We mathematically analyze the speed and spatial stationary distributions of the nodes and derive explicit expressions for the one dimensional case. In order to keep the stationary distribution through the entire simulation procedure, we provide strategies to initialize the speed, location and destination of the nodes at the beginning of the simulation. The simulation results verify the derivations and the proposed methods in this paper. This work gives a deep understanding of the properties of the Random Waypoint with Base Point mobility model and such understanding is necessary to avoid misinterpretation of the simulation results. The conclusions are of practical value for performance analysis of mobile wireless networks, especially for the catastrophe rescue scenario.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.631-634
• /
• 2004

In this paper, the Bayesian recurrent neural network (BRNN) is proposed to predict time series data. Among the various traditional prediction methodologies, a neural network method is considered to be more effective in case of non-linear and non-stationary time series data. A neural network predictor requests proper learning strategy to adjust the network weights, and one need to prepare for non-linear and non-stationary evolution of network weights. The Bayesian neural network in this paper estimates not the single set of weights but the probability distributions of weights. In other words, we sets the weight vector as a state vector of state space method, and estimates its probability distributions in accordance with the Bayesian inference. This approach makes it possible to obtain more exact estimation of the weights. Moreover, in the aspect of network architecture, it is known that the recurrent feedback structure is superior to the feedforward structure for the problem of time series prediction. Therefore, the recurrent network with Bayesian inference, what we call BRNN, is expected to show higher performance than the normal neural network. To verify the performance of the proposed method, the time series data are numerically generated and a neural network predictor is applied on it. As a result, BRNN is proved to show better prediction result than common feedforward Bayesian neural network.

• Journal of applied mathematics & informatics
• /
• v.27 no.5_6
• /
• pp.1033-1047
• /
• 2009

In this paper, we consider a m-type risk model with Markov-modulated premium rate. A integral equation for the conditional ruin probability is obtained. A recursive inequality for the ruin probability with the stationary initial distribution and the upper bound for the ruin probability with no initial reserve are given. A system of Laplace transforms of non-ruin probabilities, given the initial environment state, is established from a system of integro-differential equations. In the two-state model, explicit formulas for non-ruin probabilities are obtained when the initial reserve is zero or when both claim size distributions belong to the $K_n$-family, n $\in$ $N^+$ One example is given with claim sizes that have exponential distributions.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.30 no.9 s.252
• /
• pp.873-881
• /
• 2006

The present study investigates the pressure drop characteristics in rotating two-pass ducts. The duct has an aspect ratio (W/H) of 0.5 and a hydraulic diameter $(D_h)$ of 26.67mm. Rib turbulators are attached crossly in the four different arrangements on the leading and trailing surfaces of the test ducts. The ribs have a rectangular cross section of $2mm(e){\times}3mm(w)$ and an attack angle of $70^{\circ}C$. The pitch-to-rib height ratio (pie) is 7.5, and the rib height-to-hydraulic diameter ratio $(e/D_h)$ is 0.075. The results show that the highest pressure drop among each region appears in the turning region for the stationary case, but appears in the upstream region of the second pass for the rotating case. Effects of cross rib arrangements are almost the same in the first pass for the stationary and rotating cases. In the second pass, however, heat transfer and pressure drop are high for the cases with cross NN or PP type ribs in the stationary ducts. In the rotating ducts, they are high for the cases with cross NP or PP type ribs.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.30 no.9 s.252
• /
• pp.882-890
• /
• 2006

The present study investigates the pressure drop characteristics in rotating two-pass ducts. The duct has an aspect ratio (W/H) of 0.5 and a hydraulic diameter $(D_h)$ of 26.67mm. Rib turbulators are attached parallel in the four different arrangements on the leading and trailing surfaces of the test ducts. The ribs have a rectangular cross section of $2m(e){\times}3mm(w)$ and an attack angle of $70^{\circ}C$. The pitch-to-rib height ratio (p/e) is 7.5, and the rib height-to-hydraulic diameter ratio $(e/D_h)$ is 0.075. The results show that the highest pressure drop among each region appears in the turning region for the stationary case, but appears in the upstream region of the second pass for the rotating case. Effects of parallel rib arrangements are almost the same in the first pass for the stationary and rotating cases. In the second pass, however, heat transfer and pressure drop are high for the cases with parallel NN or PP type ribs in the stationary ducts. In the rotating ducts, they are high for the cases with parallel NN or PN type ribs.

• Communications for Statistical Applications and Methods
• /
• v.14 no.2
• /
• pp.287-299
• /
• 2007

We consider a multiclass queue where queued customers are served in their order of arrival at a rate which depends on the customer type. By using the asymptotic results obtained by Dabrowski et al. (2006) we calculate the sharp asymptotics of the stationary distribution of the number of customers of each class in the system and the distribution of the number of customers of each class when the total number of customers reaches a high level before emptying. We also obtain a fast simulation algorithm to estimate the overflow probability and compare it with the general simulation and asymptotic results.

• Journal of the Korean Statistical Society
• /
• v.35 no.1
• /
• pp.37-47
• /
• 2006

Recently, as a result of the growing interest in modeling stationary processes with discrete marginal distributions, several models for integer valued time series have been proposed in the literature. One of these models is the integer-valued autoregressive (INAR) models. However, when modeling with integer-valued autoregressive processes, the distributional properties of forecasts have been not yet discovered due to the difficulty in handling the Steutal Van Ham thinning operator 'o' (Steutal and van Ham, 1979). In this study, we derive the mean squared error of h-step-ahead prediction from a Poisson INAR(1) process, reflecting the effect of the variability of parameter estimates in the prediction mean squared error.

• Proceedings of the Korean Statistical Society Conference
• /
• 2005.11a
• /
• pp.159-165
• /
• 2005

Recently, as a result of the growing interest in modelling stationary processes with discrete marginal distributions, several models for integer valued time series have been proposed in the literature. One of theses models is the integer-valued autoregressive(INAR) models. However, when modelling with integer-valued autoregressive processes, there is not yet distributional properties of forecasts, since INAR process contain an accrued level of complexity in using the Steutal and Van Harn(1979) thinning operator 'o'. In this study, a manageable expression for the asymptotic mean square error of predicting more than one-step ahead from an estimated poisson INAR(1) model is derived. And, we present a bootstrap methods developed for the calculation of forecast interval limits of INAR(p) model. Extensive finite sample Monte Carlo experiments are carried out to compare the performance of the several bootstrap procedures.