• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.93-96
• /
• 2019

최근 딥러닝 모델을 다양한 도메인에 적용하여 뛰어난 성능을 보여주고 있다. 하지만 딥러닝 모델은 정답으로 제시된 결과가 정상적으로 예측된 결과인지, 단순히 오버피팅에 의해 예측된 결과인지를 구분하기 어렵다. 이러한 불확실성(Uncertainty)을 측정 할 수 없다는 문제점을 해결하기 위해서 본 논문에서는 베이지안 딥러닝 방법 중 하나인 변분추론(Variational Inference)과 몬테카를로 Dropout을 오픈도메인(Open-Domain) 태스크에 적용하고, 예측 결과에 대한 불확실성을 측정하여 예측결과에 영향을 주는 모델의 성능을 측정해 효과성을 보인다.

• The Korean Journal of Applied Statistics
• /
• v.10 no.2
• /
• pp.261-274
• /
• 1997

Most of the data sets to which the conventional discriminant rules have been applied contain only those which belong to one and only one class among the classes of interest. However the extension of the bivalence to multivlaence like Fuzzy concepts strongly influence the traditional view that an object must belong to only class. Thus the goal of this paper is to develop new discriminant rules which can handle the data each object of which may belong to moer than two classes with certain degrees of belongings. A calibration model is used for the relationship between the feature vector of an object and the degree of belongings and a Bayesian inference is made with the Metropolis algorithm on the degree of belongings when a feature vector of an object whose membership is unknown is given. An evalution criterion is suggested for the rules developed in this paper and comparision study is carried using two training data sets.

• 한국HCI학회:학술대회논문집
• /
• 2006.02a
• /
• pp.381-386
• /
• 2006

오늘날 분산 컴퓨팅 환경에서 운용되는 시스템이 증가함에 따라 시스템의 관리작업은 고수준(high-level)의 자동화에 대한 요구가 증가하고 있다. 이에 따라 시스템 관리방식이 전통적인 관리자 중심의 방식에서 시스템 스스로가 자신의 문제를 인식하고 상황을 분석하여 해결하는 자율 컴퓨팅 방식으로 변화하고 있으며, 이에 대한 연구가 많은 연구기관에서 다양한 방법으로 이루어지고 있다. 그러나 이러한 대부분의 기존 연구들은 문제가 발생한 이후의 치유에 주로 초점이 맞추어져 있다. 이러한 문제를 해결하기 위해서는 시스템 스스로가 동작환경을 인식하고 에러의 발생을 예측하기 위한 예측 모델이 필요하다. 따라서 본 논문에서는 자율 컴퓨팅환경에서 자가 치유를 지원하는 4가지의 예측 모델 설계 방법을 제안한다. 본 예측 모델은 ID3 알고리즘, 퍼지 추론, 퍼지 뉴럴 네트워크 그리고 베이지안 네트워크가 각 시스템 상황에 맞춰 적절하게 사용되는 방식이며, 이를 통해 보다 정확한 에러 예측이 가능해진다. 우리는 제안모델의 평가를 위해 본 예측모델을 자가치유 시스템에 적용하여 기존 연구와 예측의 효율을 비교하였으며, 그 결과를 통해 제안 모델의 유효성을 증명하였다.

• Journal of information and communication convergence engineering
• /
• v.15 no.1
• /
• pp.43-48
• /
• 2017

Movie ratings are crucial for recommendation engines that track the behavior of all users and utilize the information to suggest items the users might like. It is intuitively appealing that information about the viewing preferences in terms of movie genres is sufficient for predicting a genre of an unlabeled movie. In order to predict movie genres, we treat ratings as a feature vector, apply a Bernoulli event model to estimate the likelihood of a movie being assigned a certain genre, and evaluate the posterior probability of the genre of a given movie by using the Bayes rule. The goal of the proposed technique is to efficiently use movie ratings for the task of predicting movie genres. In our approach, we attempted to answer the question: "Given the set of users who watched a movie, is it possible to predict the genre of a movie on the basis of its ratings?" The simulation results with MovieLens 1M data demonstrated the efficiency and accuracy of the proposed technique, achieving an 83.8% prediction rate for exact prediction and 84.8% when including correlated genres.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.11 no.5
• /
• pp.139-144
• /
• 2011

Most real world system such as world economy, management, medical and engineering applications contain a series of complex phenomena. One of common methods to understand these system is to build a model and analyze the behavior of the system. As a first step, Determining the best clusters on data. As a second step, Determining the model of the cluster. In this paper, we investigated heuristic search methods for efficient clustering. It is also confirmed that the Bayesian Information Criterion more reliable than Cheeseman-Stutz ones.

• Journal of Institute of Control, Robotics and Systems
• /
• v.19 no.6
• /
• pp.513-519
• /
• 2013

The current vision-based approaches for emotion recognition, such as facial expression analysis, have many technical limitations in real circumstances, and are not suitable for applications that use them solely in practical environments. In this paper, we propose an approach for emotion recognition by combining extrinsic representations and intrinsic activities among the natural responses of humans which are given specific imuli for inducing emotional states. The intrinsic activities can be used to compensate the uncertainty of extrinsic representations of emotional states. This combination is done by using PRMs (Probabilistic Relational Models) which are extent version of bayesian networks and are learned by greedy-search algorithms and expectation-maximization algorithms. Previous research of facial expression-related extrinsic emotion features and physiological signal-based intrinsic emotion features are combined into the attributes of the PRMs in the emotion recognition domain. The maximum likelihood estimation with the given dependency structure and estimated parameter set is used to classify the label of the target emotional states.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.13 no.3
• /
• pp.186-199
• /
• 2013

We pose pattern classification as a density estimation problem where we consider mixtures of generative models under partially labeled data setups. Unlike traditional approaches that estimate density everywhere in data space, we focus on the density along the decision boundary that can yield more discriminative models with superior classification performance. We extend our earlier work on the recursive estimation method for discriminative mixture models to semi-supervised learning setups where some of the data points lack class labels. Our model exploits the mixture structure in the functional gradient framework: it searches for the base mixture component model in a greedy fashion, maximizing the conditional class likelihoods for the labeled data and at the same time minimizing the uncertainty of class label prediction for unlabeled data points. The objective can be effectively imposed as individual mixture component learning on weighted data, hence our mixture learning typically becomes highly efficient for popular base generative models like Gaussians or hidden Markov models. Moreover, apart from the expectation-maximization algorithm, the proposed recursive estimation has several advantages including the lack of need for a pre-determined mixture order and robustness to the choice of initial parameters. We demonstrate the benefits of the proposed approach on a comprehensive set of evaluations consisting of diverse time-series classification problems in semi-supervised scenarios.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.42 no.5 s.305
• /
• pp.47-54
• /
• 2005

In general, wavelet coefficients are composed of a few large coefficients and a lot of small coefficients. In this paper, we propose image denoising algorithm using Bernoulli-Gaussian mixture model based on sparse characteristic of wavelet coefficient. The Bernoulli-Gaussian mixture is composed of the multiplication of Bernoulli random variable and Gaussian mixture random variable. The image denoising is performed by using Bayesian estimation. We present an effective denoising method through simplified parameter estimation for Bernoulli random variable using local expected squared error. Simulation results show our method outperforms the states-of-art denoising methods when using orthogonal wavelets.

• The Korean Journal of Applied Statistics
• /
• v.28 no.2
• /
• pp.123-136
• /
• 2015

Science has developed with great achievements after Galileo's discovery of the law depicting a relationship between observable variables. However, many natural phenomena have been better explained by models including unobservable random effects. A mixed effect model was the first statistical model that included unobservable random effects. The importance of the mixed effect models is growing along with the advancement of computational technologies to infer complicated phenomena; subsequently mixed effect models have extended to various statistical models such as hierarchical generalized linear models. Hierarchical likelihood has been suggested to estimate unobservable random effects. Our special issue about mixed effect models shows how they can be used in statistical problems as well as discusses important needs for future developments. Frequentist and Bayesian approaches are also investigated.

• Korean Journal of Remote Sensing
• /
• v.24 no.1
• /
• pp.79-89
• /
• 2008

Irregular temporal sampling is a common feature of geophysical and biological time series in remote sensing. This study proposes an on-line system for reconstructing observation image series contaminated by noises resulted from mechanical problems or sensing environmental condition. There is also a high likelihood that during the data acquisition periods the target site corresponding to any given pixel may be covered by fog or cloud, thereby resulting in bad or missing observation. The surface parameters associated with the land are usually dependent on the climate, and many physical processes that are displayed in the image sensed from the land then exhibit temporal variation with seasonal periodicity. A feedback system proposed in this study reconstructs a sequence of images remotely sensed from the land surface having the physical processes with seasonal periodicity. The harmonic model is used to track seasonal variation through time, and a Gibbs random field (GRF) is used to represent the spatial dependency of digital image processes. The experimental results of this simulation study show the potentiality of the proposed system to reconstruct the image series observed by imperfect sensing technology from the environment which are frequently influenced by bad weather. This study provides fundamental information on the elements of the proposed system for right usage in application.