• Proceedings of the KIEE Conference
• /
• 1997.11a
• /
• pp.600-602
• /
• 1997

In this study, a hybrid neural net consisting of an Adaptive LVQ(ALVQ) algorithm and MLP is proposed to perform speaker identification task. ALVQ is a new learning procedure using adaptively feature vector sequence instead of only one feature vector in training codebooks initialized by LBG algorithm and the optimization criterion of this method is consistent with the speaker classification decision rule. ALVQ aims at providing a compressed, geometrically consistent data representation. It is fit to cover irregular data distributions and computes the distance of the input vector sequence from its nodes. On the other hand, MLP aim at a data representation to fit to discriminate patterns belonging to different classes. It has been shown that MLP nets can approximate Bayesian "optimal" classifiers with high precision, and their output values can be related a-posteriori class probabilities. The different characteristics of these neural models make it possible to devise hybrid neural net systems, consisting of classification modules based on these two different philosophies. The proposed method is compared with LBG algorithm, LVQ algorithm and MLP for performance.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06b
• /
• pp.336-338
• /
• 2012

본 연구에서는 고차 데이터 분류를 위해 순차적 베이지만 샘플링 기반의 진화연산 기법을 이용한 하이퍼네트워크 모델의 학습 알고리즘을 제시한다. 제시하는 방법에서는 모델의 조건부 확률의 사후(posterior) 분포를 최대화하도록 학습이 진행된다. 이를 위해 사전(prior) 분포를 문제와 관련된 사전지식(prior knowledge) 및 모델 복잡도(model complexity)로 정의하고, 측정된 모델의 분류성능을 우도(likelihood)로 사 용하며, 측정된 사전분포와 우도를 이용하여 모델의 적합도(fitness)를 정의한다. 이를 통해 하이퍼네트워크 모델은 고차원 데이터를 효율적으로 학습 가능할 뿐이 아니라 모델의 학습시간 및 분류성능이 개선될 수 있다. 또한 학습 시에 파라미터로 주어지던 하이퍼에지의 구성 및 모델의 크기가 학습과정 중에 적응적으로 결정될 수 있다. 제안하는 학습방법의 검증을 위해 본 논문에서는 약 25,000개의 유전자 발현정보 데이터셋에 대한 분류문제에 모델을 적용한다. 실험 결과를 통해 제시하는 방법이 기존 하이퍼네트워크 학습 방법 뿐 아니라 다른 모델들에 비해 우수한 분류 성능을 보여주는 것을 확인할 수 있다. 또한 다양한 실험을 통해 사전분포로 사용된 사전지식이 모델 학습에 끼치는 영향을 분석한다.

• Journal of Information Processing Systems
• /
• v.9 no.3
• /
• pp.395-404
• /
• 2013

In order to classify a web page as being benign or malicious, we designed 14 basic and 16 extended features. The basic features that we implemented were selected to represent the essential characteristics of a web page. The system heuristically combines two basic features into one extended feature in order to effectively distinguish benign and malicious pages. The support vector machine can be trained to successfully classify pages by using these features. Because more and more malicious web pages are appearing, and they change so rapidly, classifiers that are trained by old data may misclassify some new pages. To overcome this problem, we selected an adaptive support vector machine (aSVM) as a classifier. The aSVM can learn training data and can quickly learn additional training data based on the support vectors it obtained during its previous learning session. Experimental results verified that the aSVM can classify malicious web pages adaptively.

• Proceedings of the Korean Information Science Society Conference
• /
• 2008.06c
• /
• pp.26-30
• /
• 2008

Currently an automated methodology based on data mining techniques is presented for the prediction of customer load patterns in long duration load profiles. The proposed our approach consists of three stages: (i) data pre-processing: noise or outlier is removed and the continuous attribute-valued features are transformed to discrete values, (ii) cluster analysis: k-means clustering is used to create load pattern classes and the representative load profiles for each class and (iii) classification: we evaluated several supervised learning methods in order to select a suitable prediction method. According to the proposed methodology, power load measured from AMR (automatic meter reading) system, as well as customer indexes, were used as inputs for clustering. The output of clustering was the classification of representative load profiles (or classes). In order to evaluate the result of forecasting load patterns, the several classification methods were applied on a set of high voltage customers of the Korea power system and derived class labels from clustering and other features are used as input to produce classifiers. Lastly, the result of our experiments was presented.

• Smart Media Journal
• /
• v.9 no.3
• /
• pp.80-89
• /
• 2020

Out - of - distribution (OOD) samples are frequently encountered when deploying a classification model in plenty of real-world machine learning-based applications. Those samples are normally sampling far away from the training distribution, but many classifiers still assign them high reliability to belong to one of the training categories. In this study, we address the problem of removing OOD examples by estimating marginal density estimation using variational autoencoder (VAE). We also investigate other proper methods, such as temperature scaling, Gaussian discrimination analysis, and label smoothing. We use Chonnam National University (CNU) weeds dataset as the in - distribution dataset and CIFAR-10, CalTeach as the OOD datasets. Quantitative results show that the proposed framework can reject the OOD test samples with a suitable threshold.

• The Bulletin of The Korean Astronomical Society
• /
• v.38 no.1
• /
• pp.61.1-61.1
• /
• 2013

In this study we apply Support Vector Machine (SVM) to the prediction of geo-effective halo coronal mass ejections (CMEs). The SVM, which is one of machine learning algorithms, is used for the purpose of classification and regression analysis. We use halo and partial halo CMEs from January 1996 to April 2010 in the SOHO/LASCO CME Catalog for training and prediction. And we also use their associated X-ray flare classes to identify front-side halo CMEs (stronger than B1 class), and the Dst index to determine geo-effective halo CMEs (stronger than -50 nT). The combinations of the speed and the angular width of CMEs, and their associated X-ray classes are used for input features of the SVM. We make an attempt to find the best model by using cross-validation which is processed by changing kernel functions of the SVM and their parameters. As a result we obtain statistical parameters for the best model by using the speed of CME and its associated X-ray flare class as input features of the SVM: Accuracy=0.66, PODy=0.76, PODn=0.49, FAR=0.72, Bias=1.06, CSI=0.59, TSS=0.25. The performance of the statistical parameters by applying the SVM is much better than those from the simple classifications based on constant classifiers.

• IE interfaces
• /
• v.20 no.4
• /
• pp.448-457
• /
• 2007

The small & micro business has the characteristics of both consumer credit risk and business credit risk. In predicting the bankruptcy for small-micro businesses, the problem is that in most cases, the financial data for evaluating business credit risks of small & micro businesses are not available. To alleviate such problem, we propose a bankruptcy prediction mechanism using the credit card sales information available, because most small businesses are member store of some credit card issuers, which is the main purpose of this study. In order to perform this study, we derive some variables and analyze the relationship between good and bad signs. We employ the new statistical learning technique, support vector machines (SVM) as a classifier. We use grid search technique to find out better parameter for SVM. The experimental result shows that credit card sales information could be a good substitute for the financial data for evaluating business credit risk in predicting the bankruptcy for small-micro businesses. In addition, we also find out that SVM performs best, when compared with other classifiers such as neural networks, CART, C5.0 multivariate discriminant analysis (MDA), and logistic regression.

• International Journal of Computer Science & Network Security
• /
• v.21 no.11
• /
• pp.89-96
• /
• 2021

This work is an extension of my work presented a robust and economically efficient method for the Discrimination of four Mung-Beans [1] varieties based on quantitative parameters. Due to the advancement of technology, users try to find the solutions to their daily life problems using smartphones but still for computing power and memory. Hence, there is a need to find the best classifier to classify the Mung-Beans using already suggested features in previous work with minimum memory requirements and computational power. To achieve this study's goal, we take the experiments on various supervised classifiers with simple architecture and calculations and give the robust performance on the most relevant 10 suggested features selected by Fisher Co-efficient, Probability of Error, Mutual Information, and wavelet features. After the analysis, we replace the Artificial Neural Network and Deep learning with a classifier that gives approximately the same classification results as the above classifier but is efficient in terms of resources and time complexity. This classifier is easily implemented in the smartphone environment.

• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.637-639
• /
• 2003

As the number of spectral bands of high spectral resolution data increases, the capability to detect more detailed classes should also increase, and the classification accuracy should increase as well. Often, it is impossible to access enough training pixels for supervise classification. For this reason, the performance of traditional classification methods isn't useful. In this paper, we propose a new model for classification that operates based on decision fusion. In this classifier, learning is performed at two steps. In first step, only training samples are used and in second step, this classifier utilizes semilabeled samples in addition to original training samples. At the beginning of this method, spectral bands are categorized in several small groups. Information of each group is used as a new source and classified. Each of this primary classifier has special characteristics and discriminates the spectral space particularly. With using of the benefits of all primary classifiers, it is made sure that the results of the fused local decisions are accurate enough. In decision fusion center, some rules are used to determine the final class of pixels. This method is applied to real remote sensing data. Results show classification performance is improved, and this method may solve the limitation of training samples in the high dimensional data and the Hughes phenomenon may be mitigated.

• The Journal of the Convergence on Culture Technology
• /
• v.10 no.1
• /
• pp.617-623
• /
• 2024

Identifying emotions through speech poses a significant challenge due to the complex relationship between language and emotions. Our paper aims to take on this challenge by employing feature engineering to identify emotions in speech through a multimodal classification task involving both speech and text data. We evaluated two classifiers-Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM)-both integrated with a BERT-based pre-trained model. Our assessment covers various performance metrics (accuracy, F-score, precision, and recall) across different experimental setups). The findings highlight the impressive proficiency of two models in accurately discerning emotions from both text and speech data.