• The Korean Journal of Applied Statistics
• /
• v.33 no.2
• /
• pp.115-122
• /
• 2020

K-means clustering uses a spherical or elliptical metric to group data points; however, it does not work well for non-convex data such as the concentric circles. Spectral clustering, based on graph theory, is a generalized and robust technique to deal with non-standard type of data such as non-convex data. Results obtained by spectral clustering often outperform traditional clustering such as K-means. In this paper, we review spectral clustering and show important issues in spectral clustering such as determining the number of clusters K, estimation of scale parameter in the adjacency of two points, and the dimension reduction technique in clustering high-dimensional data.

• Progress in Medical Physics
• /
• v.15 no.4
• /
• pp.228-236
• /
• 2004

Retinal ganglion cells transmit visual scene as an action potential to visual cortex through optic nerve. Conventional recording method using single intra- or extra-cellular electrode enables us to understand the response of specific neuron on specific time. Therefore, it is not possible to determine how the nerve impulses in the population of retinal ganglion cells collectively encode the visual stimulus with conventional recording. This requires recording the simultaneous electrical signals of many neurons. Recent advances in multi-electrode recording have brought us closer to understanding how visual information is encoded by population of retinal ganglion cells. We examined how ganglion cells act together to encode a visual scene with multi-electrode array (MEA). With light stimulation (on duration: 2 sec, off duration: 5 sec) generated on a color monitor driven by custom-made software, we isolated three functional types of ganglion cell activities; ON (35.0$\pm$4.4%), OFF (31.4$\pm$1.9%), and ON/OFF cells (34.6$\pm$5.3%) (Total number of retinal pieces = 8). We observed that nearby neurons often fire action potential near synchrony (< 1 ms). And this narrow correlation is seen among cells within a cluster which is made of 6～8 cells. As there are many more synchronized firing patterns than ganglion cells, such a distributed code might allow the retina to compress a large number of distinct visual messages into a small number of ganglion cells.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.42 no.3 s.303
• /
• pp.29-38
• /
• 2005

Although people frequently rely on intuition in managing activities, they rarely use it in developing effective decision-making support systems. In this paper, we investigate and compare the correlations between such characteristics as cognition and emotion characteristics and judgmental time-series forecasting accuracy by using a self-organizing neural network, and eventually aim to help build efficient decision-making atmosphere. The neural network used in this paper employs a self-supervised adaptive algorithm, and the feature of which is that it inherently can use correlation between input vectors by exchanging information between neuron clusters in the self-organizing layer during the training. Our experiments showed that both cognition and emotion characteristics had correlations with judgmental time-series forecasting, and that cognition characteristics had larger correlation than emotion characteristics. We also found that conceptual style had larger correlation than behavioral and analytical styles, and displeasure-sleepiness style had larger correlation than pleasure-arousal style with the forecasting.

• The Journal of the Acoustical Society of Korea
• /
• v.14 no.2
• /
• pp.101-112
• /
• 1995

In order to construct a feature map-based phoneme classification system for speech recognition, two procedures are usually required. One is clustering and the other is labeling. In this paper, we present a phoneme classification system based on the Kohonen's Self-Organizing Feature Map (SOFM) for clusterer and labeler. It is known that the SOFM performs self-organizing process by which optimal local topographical mapping of the signal space and yields a reasonably high accuracy in recognition tasks. Consequently, SOFM can effectively be applied to the recognition of phonemes. Besides to improve the performance of the phoneme classification system, we propose the learning algorithm combined with the classical K-mans clustering algorithm in fine-tuning stage. In order to evaluate the performance of the proposed phoneme classification algorithm, we first use totaly 43 phonemes which construct six intra-class feature maps for six different phoneme classes. From the speaker-dependent phoneme classification tests using these six feature maps, we obtain recognition rate of $87.2\%$ and confirm that the proposed algorithm is an efficient method for improvement of recognition performance and convergence speed.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2007.10a
• /
• pp.377-382
• /
• 2007

본 논문에서는 애견 질병에 대한 전문적인 지식이 부족한 일반인들을 대상으로 자신의 애견 건강상태를 파악 할 수 있는 진단 시스템을 제안한다. 제안된 진단 시스템은 105가지 질병과 각 질병의 증상을 데이터베이스에 구축하여 입력된 증상을 통해서 애견의 질병을 도출한다. 본 논문에서는 신경망의 자율 학습 방법인 ART2 알고리즘을 적용하여 질병을 클러스터링 하고 그 결과 값인 클러스터의 출력값과 연결강도를 데이터베이스에 저장한다. 각 질병의 증상과 관련된 질의 결과를 입력 벡터로 제시하여 학습된 질병 정보와 비교하여 애견의 건강 상태를 진단한다. 애견의 건강 상태를 진단하는데 있어서 질병과 증상의 정확한 정보는 매우 중요하다. 따라서 본 논문에서는 질병과 증상의 정보를 데이터베이스로 구축하고 질병과 증상 정보를 효율적으로 관리할 수 있도록 하였다. 제안된 진단 시스템을 구현하여 수의학 전문의가 분석한 결과, 본 논문에서 제안한 시스템이 애견 질병의 보조 진단 시스템으로서의 가능성을 확인하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2002.04a
• /
• pp.557-560
• /
• 2002

본 논문에서 제안하는 분절된 화소들의 특징추출 방법은 이진화 영상에서 수직/수평 화소들의 분절점을 탐색하여 추출하는 특징 탐색기이다. 숫자의 구조적인 면을 고려하여 사소한 부분들도 명확한 특징으로 탐지하여 추출하였고, 이러한 방법은 일반적으로 사용하여지는 특징추출 방법 몇가지를 선택하여 이용하였고, 제안하는 방법과 결합하여 필기체 숫자를 인식하였다. 인식기를 구현하기 위하여 3 개층 구조를 갖는 클러스터 MLP 신경망을 사용하였다. 실험 결과 단순히 일반적인 특징만을 활용하여 얻는 인식률 보다 훨씬 향상됨을 보여주었다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2006.05a
• /
• pp.291-296
• /
• 2006

일반적으로 운송 컨테이너의 식별자들은 크기나 위치가 정형화되어 있지 않고 외부 환경으로 인한 식별자의 형태가 훼손되어 있기 때문에 일정한 규칙으로는 찾기 힘들다. 본 논문에서는 컨테이너 영상에 대해 ART2 알고리즘을 적용하여 컨테이너 영상을 양자화한다. 제안된 ART2 알고리즘 기반 양자화 기법은 컬러정보를 클러스터링 한 후, 각 클러스터의 중심 패턴을 이용하여 원 영상의 컬러정보를 분류한다. 양자화된 컨테이너 영상에서 8 방향 윤곽선 추적 알고리즘을 적용하여 개별 식별자를 추출한다. 추출된 개별 식별자는 ART2 기반 RBF 네트워크를 개선하여 인식에 적용한다. 실제 컨테이너 영상 300장에 대해 실험한 결과, 제안한 컨테이너 식별자 인식 방법의 추출 및 인식 성능이 기존의 컨테이너 식별자 인식 방법 보다 개선된 것을 확인하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.05a
• /
• pp.290-292
• /
• 2021

최근 인코더-디코더 구조의 자연어 처리모델이 활발하게 연구가 이루어지고 있다. 인코더-디코더기반의 언어모델은 특히 본문의 내용을 새로운 문장으로 요약하는 추상(Abstractive) 요약 분야에서 널리 사용된다. 그러나 기존의 언어모델은 단일 문서 및 문장을 전제로 설계되었기 때문에 기존의 언어모델에 다중 문장을 요약을 적용하기 어렵고 주제가 다양한 여러 문장을 요약하면 요약의 성능이 떨어지는 문제가 있다. 따라서 본 논문에서는 다중 문장으로 대표적이고 상품 리뷰를 워드 임베딩의 유사도를 기준으로 클러스터를 구성하여 관련성이 높은 문장 별로 인공 신경망 기반 언어모델을 통해 요약을 수행한다. 제안하는 모델의 성능을 평가하기 위해 전체 문장과 요약 문장의 유사도를 측정하여 요약문이 원문의 정보를 얼마나 포함하는지 실험한다. 실험 결과 기존의 RNN 기반의 요약 모델보다 뛰어난 성능의 요약을 수행했다.

• Journal of the Korea Society of Computer and Information
• /
• v.10 no.6 s.38
• /
• pp.93-102
• /
• 2005

This paper proposes an approach to verify signature using autonomous self-organized Neural Network Model , fused with fuzzy membership equation of fuzzy c-means algorithm, based on the features of the signature. To overcome limitations of the functional approach and Parametric approach among the conventional on-line signature recognition approaches, this Paper presents novel autonomous signature classification approach based on clustering features. Thirty-six globa1 features and twelve local features were defined, so that a signature verifying system with FE-SONN that learns them was implemented. It was experimented for total 713 signatures that are composed of 155 original signatures and 180 forged signatures yet 378 original signatures written by oneself. The success rate of this test is more than 97.67$\%$ But, a few forged signatures that could not be detected by human eyes could not be done by the system either.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.2
• /
• pp.423-428
• /
• 2017

These days, various and tons of education information are rapidly increasing and spreading due to Internet and smart devices usage. Recently, as e-Learning usage increasing, many instructors and students (learners) need to set a goal to maximize learners' result of education and education system efficiency based on big data analytics via online recorded education historical data. In this paper, the author applied Word2Vec algorithm (neural network algorithm) to find similarity among education words and classification by clustering algorithm in order to objectively recognize and analyze online recorded education historical data. When the author applied the Word2Vec algorithm to education words, related-meaning words can be found, classified and get a similar vector values via learning repetition. In addition, through experimental results, the author proved the part of speech (noun, verb, adjective and adverb) have same shortest distance from the centroid by using clustering algorithm.