• Journal of KIISE:Software and Applications
• /
• v.37 no.9
• /
• pp.706-711
• /
• 2010

In the applications of a human speech interface, reducing the error rate in recognition is the one of the main research issues. Many previous studies attempted to correct errors using post-processing, which is dependent on a manually constructed corpus and correction patterns. We propose an automatically learnable post-processing method that is independent of the characteristics of both the domain and the speech recognizer. We divide the entire post-processing task into two steps: error detection and error correction. We consider the error detection step as a classification problem for which we apply the conditional random fields (CRFs) classifier. Furthermore, we apply transformation-based learning (TBL) to the error correction step. Our experimental results indicate that the proposed method corrects a speech recognizer's insertion, deletion, and substitution errors by 25.85%, 3.57%, and 7.42%, respectively.

• The KIPS Transactions:PartB
• /
• v.12B no.4 s.100
• /
• pp.429-436
• /
• 2005

A simple stereo matching algorithm using population-based incremental learning(PBIL) is proposed in this paper to decrease the general problem of genetic algorithms, such as memory consumption and inefficiency of search. PBIL is a variation of genetic algorithms using stochastic search and competitive teaming based on a probability vector. The structure of PBIL is simpler than that of other genetic algorithm families, such as serial and parallel ones, due to the use of a probability vector. The PBIL strategy is simplified and adapted for stereo matching circumstances. Thus, gene pool, chromosome crossover, and gene mutation we removed, while the evolution rule, that fitter chromosomes should have higher survival probabilities, is preserved. As a result, memory space is decreased, matching rules are simplified and computation cost is reduced. In addition, a scheme controlling the distance of neighbors for disparity smoothness is inserted to obtain a wide-area consistency of disparities, like a result of coarse-to-fine matchers. Because of this scheme, the proposed algorithm can produce a stable disparity map with a small fixed-size window. Finally, an alterative version of the proposed algorithm without using probability vector is also presented for simpler set-ups.

• Journal of Industrial Convergence
• /
• v.20 no.5
• /
• pp.61-68
• /
• 2022

In recent years, malicious codes are being produced using the developing information and communication technology, and it is insufficient to detect them with the existing detection system. In order to accurately and efficiently detect and respond to such intelligent malicious code, an intelligent detection model is required, and in order to maximize detection performance, it is important to train with the main characteristic information set of the malicious code. In this paper, we proposed a technique for designing an intelligent detection model and generating the data required for model training as a set of key feature information through transformation, dimensionality reduction, and feature selection steps. And based on this, the main characteristic information was classified by malicious code. In addition, based on the classified characteristic information, we derived common characteristic information that can be used to analyze and detect modified or newly emerging malicious codes. Since the proposed detection model detects malicious codes by learning with a limited number of characteristic information, the detection time and response are fast, so damage can be greatly reduced and Although the performance evaluation result value is slightly different depending on the learning algorithm, it was found through evaluation that most malicious codes can be detected.

• Journal of the Institute of Convergence Signal Processing
• /
• v.24 no.2
• /
• pp.90-96
• /
• 2023

Securing large amounts of training data in deep learning neural networks, including convolutional neural networks, is of importance for avoiding overfitting phenomenon or for the excellent performance. However, securing labeled training data in deep learning neural networks is very limited in reality. To overcome this, several augmentation methods have been proposed in the literature to generate an additional large amount of training data through transformation or manipulation of the already acquired traing data. However, unlike training data such as images and texts, it is barely to find an augmentation method in the literature that additionally generates bio-signal training data for convolutional neural network based human activity recognition. Thus, this study proposes a simple but effective augmentation method of bio-signal training data for convolutional neural network based human activity recognition. The usefulness of the proposed augmentation method is validated by showing that human activity is recognized with high accuracy by convolutional neural network trained with its augmented bio-signal training data.

• Smart Media Journal
• /
• v.12 no.11
• /
• pp.77-85
• /
• 2023

After COVID-19, communication through online platforms has increased, leading to an accumulation of massive amounts of conversational text data. With the growing importance of summarizing this text data to extract meaningful information, there has been active research on deep learning-based abstractive summarization. However, conversational data, compared to structured texts like news articles, often contains missing or transformed information, necessitating consideration from multiple perspectives due to its unique characteristics. In particular, vocabulary omissions and unrelated expressions in the conversation can hinder effective summarization. Therefore, in this study, we restructured by considering the characteristics of Korean conversational data, fine-tuning a pre-trained text summarization model based on KoBART, and improved conversation data summary perfomance through a refining operation to remove redundant elements from the summary. By restructuring the sentences based on the order of utterances and extracting a central speaker, we combined methods to restructure the conversation around them. As a result, there was about a 4 point improvement in the Rouge-1 score. This study has demonstrated the significance of our conversation restructuring approach, which considers the characteristics of dialogue, in enhancing Korean conversation summarization performance.

• Journal of the Korea Society of Computer and Information
• /
• v.19 no.3
• /
• pp.25-35
• /
• 2014

This paper proposes a method of boosting-based classification for the purpose of real-time face detection. The proposed method uses depth images to ensure strong performance of face detection in response to changes in lighting and face size, and uses the depth difference feature to conduct learning and recognition through the I-MCTBoost classifier. I-MCTBoost performs recognition by connecting the strong classifiers that are constituted from weak classifiers. The learning process for the weak classifiers is as follows: first, depth difference features are generated, and eight of these features are combined to form the weak classifier, and each feature is expressed as a binary bit. Strong classifiers undergo learning through the process of repeatedly selecting a specified number of weak classifiers, and become capable of strong classification through a learning process in which the weight of the learning samples are renewed and learning data is added. This paper explains depth difference features and proposes a learning method for the weak classifiers and strong classifiers of I-MCTBoost. Lastly, the paper presents comparisons of the proposed classifiers and the classifiers using conventional MCT through qualitative and quantitative analyses to establish the feasibility and efficiency of the proposed classifiers.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.38 no.4
• /
• pp.335-341
• /
• 2010

To improve the performance of inner loop based on PD controller for a unmanned helicopter, neural networks are applied. The performance of PD controller designed on the response characteristics of error dynamics decreases because of uncertain nonlinearities of the system. The nonlinearities are decoupled to modified dynamic inversion model(MDIM) and are compensated by the neural networks. For the training of the neural networks, online weight adaptation laws which are derived from Lyapunov's direct method are used to guarantee the stability of the controller. The results of the improved performance of PD controller by neural networks are illustrated in the simulation of unmanned helicopter with nonlinearities,

• Proceedings of the Korea Information Processing Society Conference
• /
• 2008.05a
• /
• pp.455-458
• /
• 2008

단백질체학에서 2-DE는 조직내의 단백질을 규명하는 단백질 분리 기술로서 2-DE에 의하여 생성된 단백질 이미지에서 스팟 매칭을 진행하여 상이한 단백질 젤 내에 존재하는 동일한 단백질 클래스를 찾을 수 있다. 그러나 단백질 2-DE 이미지는 실험 환경의 변화에 민감하여 이미지의 위치적인 변형이나 먼지, 공기방울 등으로 인해 많은 에러 정보를 포함할 수 있다. 이러한 에러는 스팟 매칭에 치명적인 영향을 주어 낮은 정확도를 가지게 된다. 본 논문에서는 단백질 2-DE 이미지 분석을 위한 스팟 매칭에서의 정확도를 향상시키기 위하여 기준점 학습과 기준점 추출의 두 단계로 이루어진 자동화된 기준점 추출 방법을 사용하여 스팟 매칭의 정확도를 향상시킬 수 있는 최적의 기준점을 선정하는 방법을 제안하며 선정된 기준점을 기반으로 다수의 기준 이미지를 선택하여 스팟 매칭을 반복적으로 진행함으로써 확률 기반의 정확한 스팟 매칭 결과를 도출하고자 한다. 특히 데이터 마이닝 기법에서 사용되는 최소지지도 값을 적용함으로써 지지도가 높은 스팟 매칭 결과를 빈발한 스팟 매칭으로 판정한다. 제안한 스팟 매칭 정확도 향상 기법의 정확도를 평가하기 위하여 실제 단백질 2-DE 젤 이미지 데이터를 사용하여 입력 기준점의 개수와 최소 지지도의 증가에 따른 정확도의 변화를 분석하였다.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.7
• /
• pp.27-37
• /
• 2020

In this paper, we propose the methodology based on data-driven approach using Natural Language Processing and Machine Learning for classifying requirements into functional requirements and non-functional requirements. Through the analysis of the results of the requirements classification, we have learned that the trained models derived from requirements classification with data-preprocessing and classification algorithm based on the characteristics and information of existing requirements that used term weights based on TF and IDF outperformed the results that used stemming and stop words to classify the requirements into functional and non-functional requirements. This observation also shows that the term weight calculated without removal of the stemming and stop words influenced the results positively. Furthermore, we investigate an optimized method for the study of classifying software requirements into functional and non-functional requirements.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.39 no.3
• /
• pp.18-31
• /
• 2002

In this study, we introduce a concept of advanced neurofuzzy polynomial networks(ANFPN), a hybrid modeling architecture combining neurofuzzy networks(NFN) and polynomial neural networks(PNN). These networks are highly nonlinear rule-based models. The development of the ANFPN dwells on the technologies of Computational Intelligence(Cl), namely fuzzy sets, neural networks and genetic algorithms. NFN contributes to the formation of the premise part of the rule-based structure of the ANFPN. The consequence part of the ANFPN is designed using PNN. At the premise part of the ANFPN, NFN uses both the simplified fuzzy inference and error back-propagation learning rule. The parameters of the membership functions, learning rates and momentum coefficients are adjusted with the use of genetic optimization. As the consequence structure of ANFPN, PNN is a flexible network architecture whose structure(topology) is developed through learning. In particular, the number of layers and nodes of the PNN are not fixed in advance but is generated in a dynamic way. In this study, we introduce two kinds of ANFPN architectures, namely the basic and the modified one. Here the basic and the modified architecture depend on the number of input variables and the order of polynomial in each layer of PNN structure. Owing to the specific features of two combined architectures, it is possible to consider the nonlinear characteristics of process system and to obtain the better output performance with superb predictive ability. The availability and feasibility of the ANFPN are discussed and illustrated with the aid of two representative numerical examples. The results show that the proposed ANFPN can produce the model with higher accuracy and predictive ability than any other method presented previously.