• Journal of Intelligence and Information Systems
• /
• v.1 no.2
• /
• pp.17-32
• /
• 1995

본 논문에서는 추론엔진 (inference engine)내에 퍼지정보 검색부(Fuzzy Information Retrieval part)를 갖는 교통신도 제어 전문가 시스템을 제안한다. 제안하는시스템은 다양하고 복잡한 도로 상화을 고려하여 그에 따른 적절한 주기를 각 도로별로 할당함으로써 원활한 교통 흐름을 제어한다. 추론엔진내의 퍼지정보 검색부는 퍼지 삼각 논리곱을 이용하여 도로의 상황을 분석한 후 각 도로에 맞는 가장 적절한 신호주기를 생성한다.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.12 no.1
• /
• pp.7-13
• /
• 2002

Almost applications using fuzzy theory are based on the fuzzy inference. However fuzzy inference needs much time in calculation process for the fuzzy system with many input variables or many fuzzy labels defined on each variable. Inference time is dependent on the number of arithmetic Product in computation Process. Especially, the inference time is a primary constraint to fuzzy control applications using microprocessor or PC-based controller. In this paper, a simple fast fuzzy inference algorithm(FFIA), without loss of information, was proposed to reduce the inference time based on the fuzzy system with triangular membership functions in antecedent part of fuzzy rule. The proposed algorithm was induced by using partition of input state space and simple geometrical analysis. By using this scheme, we can take the same effect of the fuzzy rule reduction.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.127-128
• /
• 2020

본 논문은 딥러닝 네트워크의 압축을 위한 양자화 오프셋의 바이어스 기법을 제안한다. 양자화는 32비트 정밀도를 갖는 가중치와 활성화 데이터를 특정 비트 이하의 정수로 압축한다. 양자화는 원 데이터에 스케일과 오프셋을 더함으로써 수행되므로 오프셋을 위한 합성곱 연산이 추가된다. 본 논문에서는 입력 활성화 데이터의 양자화 오프셋과 가중치의 합성곱의 출력은 바이어스에 임베딩될 수 있음을 보여준다. 이를 통해 추론 과정 중 오프셋의 합성곱 연산을 제거할 수 있다. 실험 결과는 오프셋의 합성곱이 바이어스에 임베딩이 되더라도 영상 분류 정확도에 영향이 거의 없음을 증명한다.

• Journal of Elementary Mathematics Education in Korea
• /
• v.22 no.4
• /
• pp.385-403
• /
• 2018

Fraction division can be categorized as partitive division, measurement division, and the inverse of a Cartesian product. In the contexts of quotitive division and the inverse of a Cartesian product, the multiply-by-the-reciprocal algorithm is drawn well out. In this study, I analyze the potential and significance of the method of using $1{\div}$(divisor) as an alternative way of developing the multiply-by-the-reciprocal algorithm in the context of quotitive division. The method of using $1{\div}$(divisor) in quotitive division has the following advantages. First, by this method we can draw the multiply-by-the-reciprocal algorithm keeping connection with the context of quotitive division. Second, as in other contexts, this method focuses on the multiplicative relationship between the divisor and 1. Third, as in other contexts, this method investigates the multiplicative relationship between the divisor and 1 by two kinds of reasoning that use either ${\frac{1}{the\;denominator\;of\;the\;divisor}}$ or the numerator of the divisor as a stepping stone. These advantages indicates the potential of this method in understanding the multiply-by-the-reciprocal algorithm as the common structure of fraction division. This method is based on the dual meaning of a fraction as a quantity and the composition of times which the current elementary mathematics textbook does not focus on. It is necessary to pay attention to how to form this basis when developing teaching materials for fraction division.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.4
• /
• pp.1025-1034
• /
• 1997

Inference process of resolution-based automated reasoning easily consumes the memory of computer without giving any useful result by priducing lots of fruioless information which are not necessary for the conslusion. This paper suggests a control strategy for saving the space of computer memory and reducing the inference time. The strategy uses a restriction that comparatively irrelevant axioms do mot take pare in the resoluition. In order to analyze and determine the priorities of the input axioms of joning the inference process, the system employs the fuzzy relational products.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.3
• /
• pp.91-100
• /
• 2020

Unlike the existing Visual Question Answering(VQA) problems, the new Visual Commonsense Reasoning(VCR) problems require deep common sense reasoning for answering questions: recognizing specific relationship between two objects in the image, presenting the rationale of the answer. In this paper, we propose a novel deep neural network model, KG_VCR, for VCR problems. In addition to make use of visual relations and contextual information between objects extracted from input data (images, natural language questions, and response lists), the KG_VCR also utilizes commonsense knowledge embedding extracted from an external knowledge base called ConceptNet. Specifically the proposed model employs a Graph Convolutional Neural Network(GCN) module to obtain commonsense knowledge embedding from the retrieved ConceptNet knowledge graph. By conducting a series of experiments with the VCR benchmark dataset, we show that the proposed KG_VCR model outperforms both the state of the art(SOTA) VQA model and the R2C VCR model.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.5 no.4
• /
• pp.18-32
• /
• 1995

Fuzzy rules, developed by experts thus far, may be often inconsistent and incomplete. This paper proposes a new methodology for automatic generation of fuzzy rules which are nearly complete and not inconsistent. This is accomplished by simulating a knowledge gathering process of humans from control experiences. This method is simpler and more efficient than existing ones. It is shown through simulation that our method even generates better rules than those generated by experts, under fine tuned parameters.

• Journal of Elementary Mathematics Education in Korea
• /
• v.20 no.4
• /
• pp.521-539
• /
• 2016

The structures of partitive and quotitive division of fractions are dealt with differently, and this led to using partitive division context for helping develop invert-multiply algorithm and quotitive division for common denominator algorithm. This approach is unlikely to provide children with an opportunity to develop an understanding of common structure involved in solving different types of division. In this study, I propose two approaches, measurement approach and isomorphism approach, to develop a unifying understanding of fraction division. From each of two approaches of solving quotitive division based on proportional reasoning, I discuss an idea of constructing a measure space, unit of which is a quantity of divisor, and another idea of constructing an isomorphic relationship between the measure spaces of dividend and divisor. These ideas support invert-multiply algorithm for quotitive as well as partitive division and bring proportional reasoning into the context of fraction division. I also discuss some curriculum issues regarding fraction division and proportion in order to promote the proposed unifying understanding of partitive and quotitive division of fractions.

• Journal of Educational Research in Mathematics
• /
• v.19 no.1
• /
• pp.81-98
• /
• 2009

This study analyzed the types of quantitative reasoning and the characteristics of representation in order to figure out the characteristics of quantitative reasoning of the sixth graders. Three students who used quantitative reasoning in solving problems were interviewed in depth. Results showed that the three students used two types of quantitative reasoning, that is difference reasoning and multiplicative reasoning. They used qualitatively different quantitative reasoning, which had a great impact on their problem-solving strategy. Students used symbolic, linguistic and visual representations. Particularly, they used visual representations to represent quantities and relations between quantities included in the problem situation, and to deduce a new relation between quantities. This result implies that visual representation plays a prominent role in quantitative reasoning. This paper included several implications on quantitative reasoning and quantitative approach related to early algebra education.

• Journal of IKEEE
• /
• v.25 no.4
• /
• pp.664-671
• /
• 2021

This paper proposes a GPGPU structure that can reduce the number of operations and memory access by effectively applying a data reuse method to a convolutional neural network(CNN). Convolution is a two-dimensional operation using kernel and input data, and the operation is performed by sliding the kernel. In this case, a reuse method using an internal register is proposed instead of loading kernel from a cache memory until the convolution operation is completed. The serial operation method was applied to the convolution to increase the effect of data reuse by using the principle of GPGPU in which instructions are executed by the SIMT method. In this paper, for register-based data reuse, the kernel was fixed at 4×4 and GPGPU was designed considering the warp size and register bank to effectively support it. To verify the performance of the designed GPGPU on the CNN, we implemented it as an FPGA and then ran LeNet and measured the performance on AlexNet by comparison using TensorFlow. As a result of the measurement, 1-iteration learning speed based on AlexNet is 0.468sec and the inference speed is 0.135sec.