• The Journal of the Acoustical Society of Korea
• /
• v.20 no.8
• /
• pp.12-18
• /
• 2001

In this paper, we propose the Recurrent Neural Predictive HMM (RNPHMM). The RNPHMM is the hybrid network of the recurrent neural network and HMM. The predictive recurrent neural network trained to predict the future vector based on several last feature vectors, and defined every state of HMM. This method uses the prediction value from the predictive recurrent neural network, which is dynamically changing due to the effects of the previous feature vectors instead of the stable average vectors. The models of the RNPHMM are Elman network prediction HMM and Jordan network prediction HMM. In the experiment, we compared the recognition abilities of the RNPHMM as we increased the state number, prediction order, and number of hidden nodes for the isolated digits. As a result of the experiments, Elman network prediction HMM and Jordan network prediction HMM have good recognition ability as 98.5% for test data, respectively.

• The Journal of the Acoustical Society of Korea
• /
• v.16 no.6
• /
• pp.94-99
• /
• 1997

This paper describes a technique of probabilistic spectral subtraction which uses the knowledge of both noise and speech so as to reduce automatic speech recognition errors in noisy environments. Spectral subtraction method estimates a noise prototype in non-speech intervals and the spectrum of clean speech is obtained from the spectrum of noisy speech by subtracting this noise prototype. Thus noise can not be suppressed effectively using a single noise prototype in case the characteristics of the noise prototype are different from those of the noise contained in input noisy speech. To modify such a drawback, multiple noise prototypes are used in probabilistic subtraction method. In this paper, the probabilistic characteristics of noise and the knowledge of speech which is embedded in hidden Markov models trained in clean environments are used to suppress noise. Futhermore, dynamic feature parameters are considered as well as static feature parameters for effective noise suppression. The proposed method reduced error rates in the recognition of 50 Korean words. The recognition rate was 86.25% with the probabilistic subtraction, 72.75% without any noise suppression method and 80.25% with spectral subtraction at SNR(Signal-to-Noise Ratio) 10 dB.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.1
• /
• pp.112-119
• /
• 2010

This paper defines the probability models for determining the disparity map given stereo images and derives the methods for solving the problem, which is proven to be equivalent to an energy-based stereo matching. Under the assumptions the difference between the pixel on the left image and the corresponding pixel on the right image and the difference between the disparities of the neighboring pixels are exponentially distributed, a recursive approach for estimating the MRF regularizing parameter is proposed. Usually energy-based stereo matching methods are so sensitive to the parameter that it should be carefully determined. The proposed method alternates between estimating the parameter with the intermediate disparity map and estimating the disparity map with the estimated parameter, after computing it with random initial parameter. It is shown that the parameter estimated by the proposed method converges to the optimum and its performance can be improved significantly by adjusting the parameter and modifying the energy term.

• Journal of the Korea Society for Simulation
• /
• v.23 no.3
• /
• pp.19-25
• /
• 2014

Demand forecasting is the basis of management activities including marketing strategy. Especially, the demand of a part is remarkably important in supply chain management (SCM). In the fields of various industries, the part demand usually has the intermittent characteristic. The intermittent characteristic implies a phenomenon that there frequently occurs zero demands. In the intermittent demands, non-zero demands have large variance and their appearances also have stochastic nature. Accordingly, in the intermittent demand forecasting, it is inappropriate to apply the traditional time series models and/or cause-effect methods such as linear regression; they cannot describe the behaviors of intermittent demand. Markov bootstrap method was developed to forecast the intermittent demand. It assumes that first-order autocorrelation and independence of lead time demands. To release the assumption of independent lead time demands, this paper proposes a modified bootstrap method. The method produces the pseudo data having the characteristics of historical data approximately. A numerical example for real data will be provided as a case study.

• The Korean Journal of Applied Statistics
• /
• v.24 no.4
• /
• pp.677-693
• /
• 2011

In this paper, we consider Bayesian approaches to zero inflated Poisson model, one of the popular models to analyze zero inflated count data. To generate posterior samples, we deal with a Markov Chain Monte Carlo method using a Gibbs sampler and an exact sampling method using an Inverse Bayes Formula(IBF). Posterior sampling algorithms using two methods are compared, and a convergence checking for a Gibbs sampler is discussed, in particular using posterior samples from IBF sampling. Based on these sampling methods, a real data analysis is performed for Trajan data (Marin et al., 1993) and our results are compared with existing Trajan data analysis. We also discuss model selection issues for Trajan data between the Poisson model and zero inflated Poisson model using various criteria. In addition, we complement the previous work by Rodrigues (2003) via further data analysis using a hierarchical Bayesian model.

• Journal of KIISE:Computing Practices and Letters
• /
• v.14 no.5
• /
• pp.477-481
• /
• 2008

Place recognition is necessary for a mobile user to be provided with place-dependent information. This paper proposes real-time video based place recognition system that identifies users' current place while moving in the building. As for the feature extraction of a scene, there have been existing methods based on global feature analysis that has drawback of sensitive-ness for the case of partial occlusion and noises. There have also been local feature based methods that usually attempted object recognition which seemed hard to be applied in real-time system because of high computational cost. On the other hand, researches using statistical methods such as HMM(hidden Markov models) or bayesian networks have been used to derive place recognition result from the feature data. The former is, however, not practical because it requires huge amounts of efforts to gather the training data while the latter usually depends on object recognition only. This paper proposes a combined approach of global and local feature analysis for feature extraction to complement both approaches' drawbacks. The proposed method is applied to a mobile information system and shows real-time performance with competitive recognition result.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.2 no.2
• /
• pp.63-81
• /
• 2008

WiMAX has been introduced as a competitive alternative for metropolitan broadband wireless access technologies. It is connection oriented and it can provide very high data rates, large service coverage, and flexible quality of services (QoS). Due to the large number of connections and flexible QoS supported by WiMAX, the uplink access in WiMAX networks is very challenging since the medium access control (MAC) protocol must efficiently manage the bandwidth and related channel allocations. In this paper, we propose and investigate a cost-effective WiMAX bandwidth management scheme, named the WiMAX partial sharing scheme (WPSS), in order to provide good QoS while achieving better bandwidth utilization and network throughput. The proposed bandwidth management scheme is compared with a simple but inefficient scheme, named the WiMAX complete sharing scheme (WCPS). A maximum entropy (ME) based analytical model (MEAM) is proposed for the performance evaluation of the two bandwidth management schemes. The reason for using MEAM for the performance evaluation is that MEAM can efficiently model a large-scale system in which the number of stations or connections is generally very high, while the traditional simulation and analytical (e.g., Markov models) approaches cannot perform well due to the high computation complexity. We model the bandwidth management scheme as a queuing network model (QNM) that consists of interacting multiclass queues for different service classes. Closed form expressions for the state and blocking probability distributions are derived for those schemes. Simulation results verify the MEAM numerical results and show that WPSS can significantly improve the network’s performance compared to WCPS.

• Smart Media Journal
• /
• v.1 no.2
• /
• pp.12-22
• /
• 2012

Today we are witnessing an increasingly widespread use of cameras in our lives for video surveillance, robot vision, and mobile phones. This has led to a renewed interest in computer vision in general and an on-going boom in human activity recognition in particular. Although not particularly fancy per se, human gait is inarguably the most common and frequent action. Early on this decade there has been a passing interest in human gait recognition, but it soon declined before we came up with a systematic analysis and understanding of walking motion. This paper presents a set of DBN-based models for the analysis of human gait in sequence of increasing complexity and modeling power. The discussion centers around HMM-based statistical methods capable of modeling the variability and incompleteness of input video signals. Finally a novel idea of extending the discrete state Markov chain with a continuous density function is proposed in order to better characterize the gait direction. The proposed modeling framework allows us to recognize pedestrian up to 91.67% and to elegantly decode out two independent gait components of direction and posture through a sequence of experiments.

• Korean Journal of Fisheries and Aquatic Sciences
• /
• v.41 no.2
• /
• pp.107-112
• /
• 2008

To evaluate uncertainty and risk in biological reference points, we applied a bootstrapping method and a Bayesian procedure to estimate the related confidence intervals. Here we provide an example of the maximum sustainable yield (MSY) of turban shell, Batillus cornutus, estimated by the Schaefer and Fox models. Fitting the time series of catch and effort from 1968 to 2006 showed that the Fox model performs better than the Schaefer model. The estimated MSY and its bootstrap percentile confidence interval (CI) at ${\alpha}=0.05$ were 1,680 (1,420-1,950) tons for the Fox model and 2,170 (1,860-2,500) tons for the Schaefer model. The CIs estimated by the Bayesian approach gave similar ranges: 1,710 (1,450-2,000) tons for the Fox model and 2,230 (1,760-2,930) tons for the Schaefer model. Because uncertainty in effort and catch data is believed to be greater for earlier years, we evaluated the influence of sequentially excluding old data points by varying the first year of the time series from 1968 to 1992 to run 'backward' bootstrap resampling. The results showed that the means and upper 2.5% confidence limit (CL) of MSY varied greatly depending on the first year chosen whereas the lower 2.5% CL was robust against the arbitrary selection of data, especially for the Schaefer model. We demonstrated that the bootstrap and Bayesian approach could be useful in precautionary fisheries management, and we advise that the lower 2.5% CL derived by the Fox model is robust and a better biological reference point for the turban shells of Jeju Island.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.1
• /
• pp.61-70
• /
• 1996

According to topics or speaker's intentions in a dialogue, utterance spoken by speaker has a different sentence structure of word combinations. Based on these facts, we have proposed the statistical approach. IDT(intention decision table), which is modeling the correlations between sentence patterns and the intention. In a IDT, the sentence is splitted into 5 different factors, and the intention of a sentence is determined by the similarity between and intention and 5 factors that have represent a sentence. From the experimental results, the IDT has indicated that the prediction rate of an intention is improved 10~18% over the word-intention correlations and is enhanced 3~12% compared with the MIG(Markov intention graph) that models the intention with a transition graph for word categories in a sentence. Based on these facts, we have found that the IDT is effective method for the prediction of an intention.