• 대한음성학회지:말소리
• /
• 제52호
• /
• pp.145-160
• /
• 2004

In this paper, a new noise compensation method based on the eigenvoice framework in feature space is proposed to reduce the mismatch between training and testing environments. The difference between clean and noisy environments is represented by the linear combination of K eigenvectors that represent the variation among environments. In the proposed method, the performance improvement of speech recognition systems is largely affected by how to construct the noisy models and the bias vector set. In this paper, two methods, the one based on MAP adaptation method and the other using stereo DB, are proposed to construct the noisy models. In experiments using Aurora 2 DB, we obtained 44.86% relative improvement with eigen-environment method in comparison with baseline system. Especially, in clean condition training mode, our proposed method yielded 66.74% relative improvement, which is better performance than several methods previously proposed in Aurora project.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제3권3호
• /
• pp.285-307
• /
• 2009

This paper introduces DICE, a Domain-Independent text Classification Engine. DICE is robust, efficient, and domain-independent in terms of software and architecture. Each module of the system is clearly modularized and encapsulated for extensibility. The clear modular architecture allows for simple and continuous verification and facilitates changes in multiple cycles, even after its major development period is complete. Those who want to make use of DICE can easily implement their ideas on this test bed and optimize it for a particular domain by simply adjusting the configuration file. Unlike other publically available tool kits or development environments targeted at general purpose classification models, DICE specializes in text classification with a number of useful functions specific to it. This paper focuses on the ways to locate the optimal states of a practical text classification framework by using various adaptation methods provided by the system such as feature selection, lemmatization, and classification models.

• 반도체디스플레이기술학회지
• /
• 제10권3호
• /
• pp.75-81
• /
• 2011

This paper presents a novel method for image classification based on a hybrid genetic algorithm (GA) and support vector machine (SVM) approach which can significantly improve the classification performance for content-based image retrieval (CBIR). Though SVM has been widely applied to CBIR, it has some problems such as the kernel parameters setting and feature subset selection of SVM which impact the classification accuracy in the learning process. This study aims at simultaneously optimizing the parameters of SVM and feature subset without degrading the classification accuracy of SVM using GA for CBIR. Using the hybrid GA and SVM model, we can classify more images in the database effectively. Experiments were carried out on a large-size database of images and experiment results show that the classification accuracy of conventional SVM may be improved significantly by using the proposed model. We also found that the proposed model outperformed all the other models such as neural network and typical SVM models.

• 한국IT서비스학회지
• /
• 제14권4호
• /
• pp.121-135
• /
• 2015

Ensemble classification is to utilize multiple classifiers instead of using a single classifier. Recently ensemble classifiers have attracted much attention in data mining community. Ensemble learning techniques has been proved to be very useful for improving the prediction accuracy. Bagging, boosting and random subspace are the most popular ensemble methods. In random subspace, each base classifier is trained on a randomly chosen feature subspace of the original feature space. The outputs of different base classifiers are aggregated together usually by a simple majority vote. In this study, we applied the random subspace method to the bankruptcy problem. Moreover, we proposed a method for optimizing the random subspace ensemble. The genetic algorithm was used to optimize classifier subset of random subspace ensemble for bankruptcy prediction. This paper applied the proposed genetic algorithm based random subspace ensemble model to the bankruptcy prediction problem using a real data set and compared it with other models. Experimental results showed the proposed model outperformed the other models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권11호
• /
• pp.4557-4572
• /
• 2020

The Alaska competition provides an opportunity to study the practical problems of real-world steganalysis. Participants are required to solve steganalysis involving various embedding schemes, inconsistency JPEG Quality Factor and various processing pipelines. In this paper, we propose a method to ensemble multiple deep learning steganalyzers. We select SRNet and RESDET as our base models. Then we design a three-layers model ensemble network to fuse these base models and output the final prediction. By separating the three colors channels for base model training and feature replacement strategy instead of simply merging features, the performance of the model ensemble is greatly improved. The proposed method won second place in the Alaska 1 competition in the end.

• 한국정보시스템학회지:정보시스템연구
• /
• 제31권4호
• /
• pp.01-23
• /
• 2022

Purpose This study aims to propose a deep learning-based fake review detection model by combining authors' linguistic features and semantic information of reviews. Design/methodology/approach This study used 358,071 review data of Yelp to develop fake review detection model. We employed linguistic inquiry and word count (LIWC) to extract 24 linguistic features of authors. Then we used deep learning architectures such as multilayer perceptron(MLP), long short-term memory(LSTM) and transformer to learn linguistic features and semantic features for fake review detection. Findings The results of our study show that detection models using both linguistic and semantic features outperformed other models using single type of features. In addition, this study confirmed that differences in linguistic features between fake reviewer and authentic reviewer are significant. That is, we found that linguistic features complement semantic information of reviews and further enhance predictive power of fake detection model.

• Communications for Statistical Applications and Methods
• /
• 제30권6호
• /
• pp.551-560
• /
• 2023

Recently, there has been significant research into the recognition of human activities using three-dimensional sequential skeleton data captured by the Kinect depth sensor. Many of these studies employ deep learning models. This study introduces a novel feature selection method for this data and analyzes it using machine learning models. Due to the high-dimensional nature of the original Kinect data, effective feature extraction methods are required to address the classification challenge. In this research, we propose using the first four moments as predictors to represent the distribution of joint sequences and evaluate their effectiveness using two datasets: The exergame dataset, consisting of three activities, and the MSR daily activity dataset, composed of ten activities. The results show that the accuracy of our approach outperforms existing methods on average across different classifiers.

• 한국전자통신학회논문지
• /
• 제19권1호
• /
• pp.317-326
• /
• 2024

생체 인식은 사람의 생체적, 행동적 특징 정보를 특정 장치로 추출하여 본인 여부를 판별하는 기술이다. 생체 인식 분야에서 생체 특성 위조, 복제, 해킹 등 사이버 위협이 증가하고 있다. 이에 대응하여 보안 시스템이 강화되고 복잡해지며, 개인이 사용하기 어려워지고 있다. 이를 위해 다중 생체 인식 모델이 연구되고 있다. 기존 연구들은 특징 융합 방법을 제시하고 있으나, 특징 융합 방법 간의 비교는 부족하다. 이에 본 논문에서는 지문, 얼굴, 홍채 영상을 이용한 다중 생체 인식 모델의 융합 방법을 비교 평가했다. 특징 추출을 위해VGG-16, ResNet-50, EfficientNet-B1, EfficientNet-B4, EfficientNet-B7, Inception-v3를 사용했으며, 특성융합을 위해 'Sensor-Level', 'Feature-Level', 'Score-Level', 'Rank-Level' 융합 방법을 비교 평가했다. 비교평가결과 'Feature-Level' 융합 방법에서 EfficientNet-B7 모델이 98.51%의 정확도를 보이며 높은 안정성을 보였다. 그러나 EfficietnNet-B7모델의 크기가 크기 때문에 생체 특성 융합을 위한 모델 경량화 연구가 필요하다.

• Communications for Statistical Applications and Methods
• /
• 제9권3호
• /
• pp.577-594
• /
• 2002

For the estimation and test of long memory feature in volatilities of stock indices and individual companies semiparametric approach, Geweke and Porter-Hudak (1983), is employed. Empirical study supports the strong evidence of volatility persistence in Korean stock market. Most of indices and individual companies have the feature of long term dependence of volatility. Hence the short memory models are unable to explain the volatilities in Korean stock market.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2008년도 하계종합학술대회
• /
• pp.759-760
• /
• 2008

In this paper, We propose the feature-based 3D model Retrieval System. 3D models are represented as triangle meshes. A first simple feature vector can be calculated from hull. After looking for meshes intersected with the hull, we compute the curvature of meshes. These curvature are used as the model descriptor.