CNN의 SoftMax 연산을 위한 연속 근사 방식의 로그 연산 회로

강형주 한국정보통신학회 2021 한국정보통신학회논문지 Vol.25 No.2

In a CNN for image classification, a SoftMax layer is usually placed at the end. The exponentinal and logarithmic operations in the SoftMax layer are not adequate to be implemented in an accelerator circuit. The operations are usually implemented with look-up tables, and the exponential operation can be implemented in an iterative method. This paper proposes a successive approximation method to calculate a logarithm to remove a very large look-up table. By substituing the large table with two very small tables, the circuit can be reduced much. The experimental results show that the 85% area reduction can be reached with a small error degradation.

SSD와 SSDLite의 비교를 통한 객체 검출 최적화

강형주 한국정보통신학회 2023 한국정보통신학회논문지 Vol.27 No.3

Convolutional Neural Networks are being widely used in computer vision tasks including object detection. For resource-limited environments, the combination of MobileNetV2 and SSDLite is a feasible solution, but the structure of SSDLite has not been well analyzed. This paper compares the structure of SSDLite with that of the previous version, SSD (Single-Shot multibox Detector), finding out possible variations. With the application of the variations, the experimental results shows that SSDLite can be improved by enhancing the capability of detecting small objects. The MS COCO AP can be raised by 0.6% point with complexity increased a little.

Retiming을 이용한 Symbolic Model Checking 성능 향상에 관한 연구

강형주,Kang, Hyeong-Ju 한국정보통신학회 2010 한국정보통신학회논문지 Vol.14 No.10

This paper presents an application of retiming to model checking, a branch of formal verification. Retiming can change the transition relation of a circuit without changing its input-output behaviour by relocating its registers. With the retiming, a given circuit can have a different structure more adequate for model checking. This paper proposes a cost function to reflect the number of registers and the characteristic of its transition relation and develops a heuristic annealing algorithm to search efficiently the circuit structures obtained by retiming. Experimental results show that the proposed method can improve the model checking performance. 이 논문에서는 형식 검증(formal verification)의 한 분야인 모델 검증(model checking)에 재타이밍(retiming) 기법을 적용하는 방법에서 대해 연구하였다. 재타이밍은 주어진 회로의 레지스터들을 재배치함으로써, 입출력 동작을 바꾸지 않으면서 전이 관계(transition relation)을 변환할 수 있는 기법이다. 이러한 재타이밍을 이용하면 모델 검증을 더 효율적으로 수행하도록 회로를 바꿀 수 있다. 이 논문에서는, 레지스터의 개수와 전이 관계의 특성을 반영한 cost 함수를 제안하고, 재타이밍으로 얻을 수 있는 회로 구조들을 효율적으로 탐색하는 heuristic annealing 알고리즘을 개발한다. 제안된 방법이 모델 검증의 성능을 향상시킬 수 있음을 실험 결과를 통해 보여주었다.

Low-Complexity Multi-size Cyclic-Shifter for QC-LDPC Codes

강형주,양병도 한국전자통신연구원 2017 ETRI Journal Vol.39 No.3

The decoding process of a quasi-cyclic low-density parity check code requires a unique type of rotator. These rotators, called multi-size cyclic-shifters (MSCSs), rotate input data with various sizes, where the size is the amount of data to be rotated. This paper proposes a low-complexity MSCS structure for the case when the sizes have a nontrivial common divisor. By combining the strong points of two previous structures, the proposed structure achieves the smallest area. The experimental results show that the area reduction was more than 14.7% when the proposed structure was applied to IEEE 802.16e as an example.

농어촌지역 성당 내 성체조배실의 건축적 특성에 관한 연구 - 광주·전남지역 천주교성당의 성체조배실을 중심으로 건축 요소 분석 -

강형주,신웅주 한국농촌건축학회 2021 농촌건축 : 한국농촌건축학회논문집 Vol.23 No.4

Religious architecture in the Gwangju and Jeonam regions played the role of local culture, and in particular, cathedral architecture maintains a consistent religious context and seeks to integrate with the local community. This study is to investigate and analyze the overall data to find out the characteristics and meaning of the Eucharistic adoration, which played a central role in the religious aspect and the space in the cathedral, and used it as basic data for the study of local cathedral architecture. The Eucharistic adoration was activated starting from the time when the Eucharist storage room was placed in the Middle Ages, and in 1979, when Pope John Paul II published a letter, and Bishop Na Gilmo of Incheon introduced the Eucharist, the Eucharistic adoration movement was activated in parishes across the country. However, regarding the Eucharistic adoration, a way to use its spatial meaning and religious use is still unknown, even to its believers. The Eucharistic adoration of the cathedral contains the main spiritual function of managing faith rather than the functions of mission, fellowship, and office work. An approach which composes the liturgical spaces in various architectural planning methods is necessary, but the fundamental and religious meaning must not be abandoned.

고용량의 Methotrexate 투여 후 발생한 급성신부전증에서 시행한 혈장교환법 1례

강형주,한혜원,김명철,박진노,진동찬,홍영선,이승구,김춘추,이경식 大韓血液學會 2001 大韓血液學會誌 Vol.36 No.3

ResNet-50 합성곱 신경망을 위한 고정 소수점 표현 방법

강형주,Kang, Hyeong-Ju 한국정보통신학회 2018 한국정보통신학회논문지 Vol.22 No.1

최근 합성곱 신경망은 컴퓨터 비전에 관련된 여러 분야에서 높은 성능을 보여 주고 있으나 합성곱 신경망이 요구하는 많은 연산양은 임베디드 환경에 도입되는 것을 어렵게 하고 있다. 이를 해결하기 위해 ASIC이나 FPGA를 통한 합성곱 신경망의 구현에 많은 관심이 모이고 있고, 이러한 구현을 위해서는 효율적인 고정 소수점 표현이 필요하다. 고정 소수점 표현은 ASIC이나 FPGA에서의 구현에 적합하나 합성곱 신경망의 성능이 저하될 수 있는 문제가 있다. 이 논문에서는 합성곱 계층과 배치(batch) 정규화 계층에 대해 고정 소수점 표현을 분리해서, ResNet-50 합성곱 신경망의 합성곱 계층을 표현하기 위해 필요한 비트 수를 16비트에서 10비트로 줄일 수 있게 하였다. 연산이 집중되는 합성곱 계층이 더 간단하게 표현되므로 합성곱 신경망 구현이 전체적으로 더 효율적으로 될 것이다. Recently, the convolutional neural network shows high performance in many computer vision tasks. However, convolutional neural networks require enormous amount of operation, so it is difficult to adopt them in the embedded environments. To solve this problem, many studies are performed on the ASIC or FPGA implementation, where an efficient representation method is required. The fixed-point representation is adequate for the ASIC or FPGA implementation but causes a performance degradation. This paper proposes a separate optimization of representations for the convolutional layers and the batch normalization layers. With the proposed method, the required bit width for the convolutional layers is reduced from 16 bits to 10 bits for the ResNet-50 neural network. Since the computation amount of the convolutional layers occupies the most of the entire computation, the bit width reduction in the convolutional layers enables the efficient implementation of the convolutional neural networks.

Mission Management Technique for Multi-sensor-based AUV Docking

강형주,Gun Rae Cho,김민규,이문직,이계홍,김호성,이한솔,이권수 한국해양공학회 2022 韓國海洋工學會誌 Vol.36 No.3

This study presents a mission management technique that is a key component of underwater docking system used to expand the operating range of autonomous underwater vehicle (AUV). We analyzed the docking scenario and AUV operating environment, defining the feasible initial area (FIA) level, event level, and global path (GP) command to improve the rate of docking success and AUV safety. Non-holonomic constraints, mounted sensor characteristic, AUV and mission state, and AUV behavior were considered. Using AUV and docking station, we conducted experiments on land and at sea. The first test was conducted on land to prevent loss and damage of the AUV and verify stability and interconnection with other algorithms; it performed well in normal and abnormal situations. Subsequently, we attempted to dock under the sea and verified its performance; it also worked well in a sea environment. In this study, we presented the mission management technique and showed its performance. We demonstrated AUV docking with this algorithm and verified that the rate of docking success was higher compared to those obtained in other studies.

CNN에서 입력 최댓값을 이용한 SoftMax 연산 기법

강형주 한국정보통신학회 2022 한국정보통신학회논문지 Vol.26 No.2

A convolutional neural network(CNN) is widely used in the computer vision tasks, but its computing power requirement needs a design of a special circuit. Most of the computations in a CNN can be implemented efficiently in a digital circuit, but the SoftMax layer has operations unsuitable for circuit implementation, which are exponential and logarithmic functions. This paper proposes a new method to integrate the exponential and logarithmic tables of the conventional circuits into a single table. The proposed structure accesses a look-up table (LUT) only with a few maximum values, and the LUT has the result value directly. Our proposed method significantly reduces the space complexity of the SoftMax layer circuit implementation. But our resulting circuit is comparable to the original baseline with small degradation in precision.

여대생을 위한 인터넷용 영양 정보 제공 및 상담 프로그램 개발

강형주,김경원 대한지역사회영양학회 2000 대한지역사회영양학회지 Vol.5 No.4