• Journal of Convergence for Information Technology
• /
• v.10 no.8
• /
• pp.23-34
• /
• 2020

Recently, the artificial intelligence deep learning field has been hard to commercialize due to the high computing power and the price problem of computing resources. In this paper, we apply a double pruning techniques to evaluate the performance of the in-depth neural network and various datasets. Double pruning combines basic Network-slimming and Parameter-prunning. Our proposed technique has the advantage of reducing the parameters that are not important to the existing learning and improving the speed without compromising the learning accuracy. After training various datasets, the pruning ratio was increased to reduce the size of the model.We confirmed that MobileNet-V3 showed the highest performance as a result of NetScore performance analysis. We confirmed that the performance after pruning was the highest in MobileNet-V3 consisting of depthwise seperable convolution neural networks in the Cifar 10 dataset, and VGGNet and ResNet in traditional convolutional neural networks also increased significantly.

• Journal of KIISE:Software and Applications
• /
• v.31 no.12
• /
• pp.1569-1580
• /
• 2004

One commonly used approach to deal with uncertainty is Bayesian network which represents joint probability distributions of domain. There are some attempts to team the structure of Bayesian networks automatically and recently many researchers design structures of Bayesian network using evolutionary algorithm. However, most of them use the only one fittest solution in the last generation. Because it is difficult to combine all the important factors into a single evaluation function, the best solution is often biased and less adaptive. In this paper, we present a method of generating diverse Bayesian network structures through fitness sharing and combining them by Bayesian method for adaptive inference. In order to evaluate performance, we conduct experiments on learning Bayesian networks with artificially generated data from ASIA and ALARM networks. According to the experiments with diverse conditions, the proposed method provides with better robustness and adaptation for handling uncertainty.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.7
• /
• pp.972-980
• /
• 2022

The butterfly species recognition technology based on machine learning using images has the effect of reducing a lot of time and cost of those involved in the related field to understand the diversity, number, and habitat distribution of butterfly species. In order to improve the accuracy and time efficiency of butterfly species classification, various features used as the inputs of machine learning models have been studied. Among them, branch length similarity(BLS) entropy or color intensity entropy methods using the concept of entropy showed higher accuracy and shorter learning time than other features such as Fourier transform or wavelet. This paper proposes a feature extraction algorithm using RGB color intensity entropy for butterfly color images. In addition, we develop butterfly recognition systems that combines the proposed feature extraction method with representative ensemble models and evaluate their performance.

• Journal of The Korean Association of Information Education
• /
• v.26 no.3
• /
• pp.167-174
• /
• 2022

In blended learning, which combines school and online learning, unidirectionalness and disconnection of interaction are pointed out as the reason for widening the educational gap. This study conducted focus group interviews with teachers, students, and parents to develop a blended learning model to enhance the effectiveness of online learning. In order to increase the interaction between teachers and students online, the EduCare model considering care factors was proposed. Six practical models were proposed for the blended EduCare model. Not only interaction can be maximized, but learning reality and psychological well-being, social presence can be improved, and the effectiveness of online learning can be cultivated positively.

• Journal of the Korea Society of Computer and Information
• /
• v.19 no.9
• /
• pp.33-38
• /
• 2014

The FCM based hybrid RBF network is a heterogeneous learning network model that applies FCM algorithm between input and middle layer and applies Max_Min algorithm between middle layer and output. The Max-Min neural network uses winner nodes of the middle layer as input but shows inefficient learning in performance when the input vector consists of too many patterns. To overcome this problem, we propose a dynamic learning rate control based on fuzzy logic. The proposed method first classifies accurate/inaccurate class with respect to the difference between target value and output value with threshold and then fuzzy membership function and fuzzy decision logic is designed to control the learning rate dynamically. We apply this proposed RBF network to the character recognition problem and the efficacy of the proposed method is verified in the experiment.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.12
• /
• pp.1832-1837
• /
• 2022

In the SW education of non-majors who encounter liberal arts education experience difficulties in the SW development environment and understanding they encounter for the first time, relevance to their major, and convergence thinking ability. In order to compensate for the difficulties of non-major learners in liberal arts education, a relatively easily accessible software was used to utilize a demonstration-oriented model that can be applied to beginners in SW education. In order to understand the logical flow of applications and problem solving used in real life, we proposed a convergence SW teaching method that combines repeated implementation through demonstration by the instructor and imitation of the learner, and learning indicators to increase the learning satisfaction and achievement of the learner. In the experiment applying the teaching and learning method proposed in this paper, meaningful results were shown when evaluating the learning effect, academic achievement, learning satisfaction, and teaching and learning method aspects of SW education.

• Korean Journal of Cognitive Science
• /
• v.23 no.3
• /
• pp.295-322
• /
• 2012

In this paper, we cluster similar word senses applying vector space model and HAL (Hyperspace Analog to Language). HAL measures corelation among words through a certain size of context (Lund and Burgess 1996). The similarity measurement between a word pair is cosine similarity based on the vector space model, which reduces distortion of space between high frequency words and low frequency words (Salton et al. 1975, Widdows 2004). We use PCA (Principal Component Analysis) and SVD (Singular Value Decomposition) to reduce a large amount of dimensions caused by similarity matrix. For sense similarity clustering, we adopt supervised and non-supervised learning methods. For non-supervised method, we use clustering. For supervised method, we use SVM (Support Vector Machine), Naive Bayes Classifier, and Maximum Entropy Method.

• The KIPS Transactions:PartB
• /
• v.14B no.7
• /
• pp.523-530
• /
• 2007

This paper proposes a new stochastic optimization algorithm for hidden Markov models (HMMs) used as a recognizer of automatic lipreading. The proposed method combines a global stochastic optimization method, the simulated annealing technique, and the local optimization method, which produces fast convergence and good solution quality. We mathematically show that the proposed algorithm converges to the global optimum. Experimental results show that training HMMs by the method yields better lipreading performance compared to the conventional training methods based on local optimization.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.147-150
• /
• 2018

목적 지향 대화 시스템은 자연어 이해, 대화 관리자, 자연어 생성과 같은 세분화 모델들의 결합으로 이루어져있어 하위 모델에 대한 오류 전파에 취약하다. 이러한 문제점을 해결하기 위해 자연어 이해 모델과 대화 관리자를 하나의 네트워크로 구성하고 오류에 강건한 심층 Q 네트워크를 제안한다. 본 논문에서는 대화의 전체 흐름을 파악 할 수 있는 순환 신경망인 LSTM에 심층 Q 네트워크 적용한 심층 순환 Q 네트워크 기반 목적 지향 대화 시스템을 제안한다. 실험 결과, 제안한 심층 순환 Q 네트워크는 LSTM, 심층 Q 네트워크보다 각각 정밀도 1.0%p, 6.7%p 높은 성능을 보였다.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2004.10a
• /
• pp.463-466
• /
• 2004

본 논문에서는 각 입력 변수에 대하여 퍼지 공간을 분할한 퍼지 집합 기반 퍼지 추론 시스템을 제안한다. 퍼지 모델은 주로 경험적 방법에 의해 추출되기 때문에 보다 구체적이고 체계적인 방법에 의한 동정 및 최적화 쥘 필요성이 요구된다. 정보 granules는 근접성, 유사성 또는 기능성 등의 기준에 의해 서로 결합된 물체(특히, 데이터 점)의 연결된 모임으로 간주된다. 정보 데이터의 특성을 살리기 위해 HCM 클러스터링 방법에 의한 중심71을 이용하여 각 입력 변수에 대한 퍼지 집합 기반 전반부/후반부 구조 및 파라미터를 동정한다. 퍼지 추론 방법은 간략 및 선형 퍼지 추론을 수행하며 삼각형 멤버쉽 함수를 사용한다. 구축된 퍼지 모델은 유전자 알고리즘을 이용하여 전반부 파라미터를 최적으로 동정하며, 학습 및 테스트 데이터의 성능 결과의 상호균형을 얻기 위한 하중값을 가진 성능지수를 사용하여 근사화와 예측성능의 향상을 꾀한다. 또한, 제안된 퍼지 모델은 수치적인 예를 통하여 성능을 평가한다.