• Journal of KIISE
• /
• v.44 no.12
• /
• pp.1282-1289
• /
• 2017

Since the study of deep convolutional neural network became prevalent, one of the important discoveries is that a feature map from a convolutional network can be extracted before going into the fully connected layer and can be used as a saliency map for object detection. Furthermore, the model can use features from each different layer for accurate object detection: the features from different layers can have different properties. As the model goes deeper, it has many latent skip connections and feature maps to elaborate object detection. Although there are many intermediate layers that we can use for semantic segmentation through skip connection, still the characteristics of each skip connection and the best skip connection for this task are uncertain. Therefore, in this study, we exhaustively research skip connections of state-of-the-art deep convolutional networks and investigate the characteristics of the features from each intermediate layer. In addition, this study would suggest how to use a recent deep neural network model for semantic segmentation and it would therefore become a cornerstone for later studies with the state-of-the-art network models.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.23 no.3
• /
• pp.232-237
• /
• 2013

The aim of this study is to structuralize a model of the factors determining the quality of nursing care perceived by the director or manager of a long-term care facilities (hospitalization of patients) using FSM(Fuzzy Structural Modeling), employed in structuralizing social systems. The results were as follows: first, quality in the top tier was shown to be connected with job commitment, commitment to the organization, work experience, care skills, knowledge about the elderly, training and education, which are factors in the middle tier; and second, the structure of the middle tier (job commitment, commitment to the organization, work experience, care skills, knowledge about the elderly, training and education) either showed a connection with the lower tier, which includes employment type, job satisfaction, leadership, relationship with users and workplace relationships, or showed a connection among the factors within. These results confirmed the following: first, care skills and knowledge about the elderly, which demonstrate the job expertise of caregivers, showed a connection with service quality based on work experience; second, job commitment in the middle tier was observed to affect various factors in the same tier such as care skills, knowledge about the elderly, training and education amongst others, and it was determined that it is an important determining factor in service quality. Lastly, a meaningful result was shown in relation to leadership. The leadership skills of the director of the facilities had a connection with the care caregivers' commitment to the organization, which had a connection with service quality. This structure showed the kind of role the director must play in order to improve service quality.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09b
• /
• pp.91-94
• /
• 2003

본 논문에서는 RBF 네트워크의 중간층과 출력층 사이의 연결강도를 효율적으로 조정하기 위해 퍼지 논리 시스템을 이용하여 학습률과 모멘텀을 동적으로 조정하는 개선된 RBF 네트워크를 제안한다. 입력층과 중간층 사이의 학습 구조로 ART2를 적용하고 중간층과 출력층 사이의 연결 강도 조정 방법으로는 제안된 학습률 자동 조정 방식을 적용한다. 제안된 방법의 학습 성능을 평가하기 위해 기존의 delta-bar-delta 알고리즘, 기존의 ART2 기반의 RBF 네트워크와 비교 분석한 결과, 제안된 방법이 학습 속도와 수렴성에서 개선된 것을 확인하였다.

• Proceedings of the Acoustical Society of Korea Conference
• /
• autumn
• /
• pp.71-76
• /
• 1999

본 논문에서 제안한 예측형 RBFN(Radial Basis Function Network)은 HMM과 신경망을 결합한 하이브리드 구조이다. 이 신경망은 HMM으로 추정한 확률분포 파라미터를 사용하여 중간층의 활성화 함수의 출력을 결정하고, 중간층과 출력층의 연결강도만 네트워크 내에서 학습한다. 그리고 HMM으로 추정한 확률분포 파라미터는 두 가지 방법으로 예측형 RBFN에 이용하였다. 첫 번째는 HMM의 각 상태의 혼합수 만큼의 중간층 유니트를 주는 방법이고, 두 번째는 HMM의 혼합수$\times$출력분포수 만큼의 중간층 유니트를 주는 방법이다. 실험결과, 예측형 RBFN은 다른 방법들의 결과보다 $4.5\~6.5\%$ 저하된 결과를 보였지만 다른 신경망에 비해서 학습 반복 횟수를 작게할 수 있었으며 전체 학습시간을 대폭 단축할 수 있었다.

• The Journal of the Korea Contents Association
• /
• v.4 no.3
• /
• pp.99-105
• /
• 2004

This paper proposes optimal learning rates in the gradient descent training of multilayer perceptrons, which are a separate learning rate for weights associated with each neuron and a separate one for assigning virtual hidden targets associated with each training pattern Effectiveness of the proposed error function was demonstrated for a handwritten digit recognition and an isolated-word recognition tasks and very fast learning convergence was obtained.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.05a
• /
• pp.511-514
• /
• 2019

인공지능이 정형화된 수치 데이터뿐만 아니라 비정형 데이터까지도 인식해야하는 시대가 왔다. 보안 분야 이외에도 사회 전반에서 숫자 인식을 활용하고 점차 확대되고 있다. 숫자인식을 위해 인공신경망을 이용하였다. 인공신경망은 입력 층, 중간 층, 출력 층으로 이루어져 있다. 각 층은 노드와 노드들을 연결하는 가중치로 구성되어 있다. data set을 입력 값으로 하여 각각의 가중치를 곱한다. 오차역전파법을 이용하여 가중치 값을 갱신한다. 갱신하는 과정에서 학습률과 가중치 조정을 통해 결과 값의 정확도를 연구한다. 궁극적으로 학습된 data set과 인공신경망 알고리즘을 이용하여 손 글씨로 된 숫자를 인식한다. 실험에서 학습률과 중간층의 노드 개수를 조정하여 인식률을 높여간다.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.26 no.6
• /
• pp.458-463
• /
• 2016

In this paper, the fuzzy structure model for the consciousness structure of social workers related to the elder abuse factors was derived and analyzed. The characteristics of the model was obtained as follows. First, the elder abuse behavior at the top layer was related to the attitude of the elderly and the work overload of social workers. Second, the attitude of the elderly and the work overload of social workers at the middle layer were related to the personality of social worker, the physical and mental dependency of client, and the personality of client. Third, the personality of social worker, the knowledge of the elderly, the personality of client, and the physical and psychological dependence of the client affected directly the elder abuse behavior without going through the middle layer. Fourth, the work overload of social workers at the middle layer was affected the attitude of the elderly. Finally, the age of social workers, the working image, the job training, and provision of punishment to the social workers were the isolated layer, in which the relationship between the elder abuse behavior and related factors was not found.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.13 no.6
• /
• pp.1200-1207
• /
• 2009

The GMDH(Group Method of Data Handling) algorithm can be used to model the complex nonlinear systems. The traditional GMDH algorithm produces the output of the system model in the output layer through the input layer and the intermediate layers as the prescribed process. The outputs of each layer are produced only by the outputs of the former layer. However among the inputs there may be the inputs which can influence the modeling result more than the other inputs. Therefore in this paper the method which improve the modeling capability by interlayer connection of more influential inputs is proposed. The capability improvement of the proposed algorithm compared to the traditional algorithm is verified through computer simulation.

• Magazine of the Korean Society of Agricultural Engineers
• /
• v.38 no.5
• /
• pp.47-54
• /
• 1996

본 연구에서는 지표면유출과 중간유출의 수문학적과정을 함께 모의발생 시키는 합성 유역모델이 제시되었다. 본 모델은 디지털지형모델과 상호 연결되도록 하였으며 지형이 복잡한 지역에서도 유출이 시간과 공간적으로 누가계산되어 이 분야의 조사연구에 필요한 정보를 제공할 수 있다. 본모델을 이용 유역의 불투수층 위에 분포해있는 토양의 중간계층과 토양수분의 계산 및 침투/용탈의 과정을 모의 발생시킬 수 있다.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.17 no.6 s.109
• /
• pp.529-537
• /
• 2006

Multi-layer PCBs are of ien used In compact microwave circuit design as density of PCB layout is increased. In this paper, the crosstalk between coupled lines connected with vias in a 4-layer PCB is investigated theoretically based on the circiuit-concept approach. Coupled lines connected with vias in a 4-layer PCB are divided into three sections, which are coupled microstrip lines and upper via section, center via section, and lower via and coupled microstrip lines section, respectively. Each section is represented by ABCD matrix. By cascading these three ABCD matrices crosstalk between coupled lines connected with vias in a 4-layer PCB is approximately calculated. The validity of this theoretical approach is verified by comparing the calculated results with the simulated ones using HFSS.