• Proceedings of the Korea Information Processing Society Conference
• /
• 2012.11a
• /
• pp.1728-1731
• /
• 2012

본 논문에서는 부모 개체의 해밍 거리에 기반하여 선택적 변이연산을 적용한 유전알고리즘을 제안한다. 유전자 형이 매우 유사한 개체들 간의 유전연산은 알고리즘의 탐색성능을 저하시키고 조기 수렴의 가능성을 증가시킨다. 본 논문에서는 이러한 현상을 극복하기 위하여, 교차연산 시 선택된 두 부모 개체간의 해밍 거리에 따라 그 값이 낮으면 교차연산 후 생성된 두 자식 개체 중 한쪽에게 높은 변이확률을 적용하고 다른 한쪽 자식은 부모와 비슷한 유전자 형으로 탐색을 계속하게 하여 조기 수렴을 방지하면서 해집단의 다양성 유지 기능을 향상 시켰다. 제안한 유전 알고리즘을 다차원 배낭 문제에 적용한 결과, 같은 조건에서 단순 유전 알고리즘(SGA) 보다 향상된 탐색 성능을 보여주었다.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.19 no.4
• /
• pp.161-166
• /
• 2018

In this paper, we propose a method for the random selection of pooling operations for the regularization and reduction of cross validation in convolutional neural networks. The pooling operation in convolutional neural networks is used to reduce the size of the feature map and for its shift invariant properties. In the existing pooling method, one pooling operation is applied in each pooling layer. Because this method fixes the convolution network, the network suffers from overfitting, which means that it excessively fits the models to the training samples. In addition, to find the best combination of pooling operations to maximize the performance, cross validation must be performed. To solve these problems, we introduce the probability concept into the pooling layers. The proposed method does not select one pooling operation in each pooling layer. Instead, we randomly select one pooling operation among multiple pooling operations in each pooling region during training, and for testing purposes, we use probabilistic weighting to produce the expected output. The proposed method can be seen as a technique in which many networks are approximately averaged using a different pooling operation in each pooling region. Therefore, this method avoids the overfitting problem, as well as reducing the amount of cross validation. The experimental results show that the proposed method can achieve better generalization performance and reduce the need for cross validation.

• Proceedings of the Korean Information Science Society Conference
• /
• 2003.10b
• /
• pp.823-825
• /
• 2003

시간이 흐름에 따라 생화학 시스템이 변화하는 것을 기록한 데이터로부터 이 시스템의 상태 전이 및 시스템을 구성하는 각 생화학 물질간의 관계를 모델링하기 위한 방법으로 S-tree 구조를 제안한다. 이것은 주로 생화학 시스템의 동적 특성을 모델링 하기 위하여 연구되어 온 S-system을 나무 구조로 표현한 것이다. 본 논문에서는 진화 연산을 통해 주어진 시계열 데이터를 잘 설명하는 S-tree의 구조 및 그 변수들을 동시에 효과적으로 탐색하는 방법을 개발하였다. 이 방법에서는 구조 탐색을 위해 유전 프로그래밍(genetic programming)에서 사용되어 온 나무 구조의 교차 및 돌연변이 연산과 더불어 다양한 형태의 구조 탐색 연산자들을 도입하였고, 또한 동시에 알맞은 변수 값들을 찾기 위하여 확률적 돌연변이 연산을 통한 언덕 오르기(hill-climbing)를 수행한다. 제안된 방법을 효모의 혐기성 발효 데이터에 적용한 결과 주어진 시스템을 성공적으로 모델링할 수 있었다.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.8
• /
• pp.371-380
• /
• 2023

Relation extraction is to extract relationships between named entities from text. Traditionally, relation extraction methods only extract relations between predetermined subject and object entities. However, in end-to-end relation extraction, all possible relations must be extracted by considering the positions of the subject and object for each pair of entities, and so this method uses time and resources inefficiently. To alleviate this problem, this paper proposes a method that sets directions based on the positions of the subject and object, and extracts relations according to the directions. The proposed method utilizes existing relation extraction data to generate direction labels indicating the direction in which the subject points to the object in the sentence, adds entity position tokens and entity type to sentences to predict the directions using a pre-trained language model (KLUE-RoBERTa-base, RoBERTa-base), and generates representations of subject and object entities through probabilistic crossover operation. Then, we make use of these representations to extract relations. Experimental results show that the proposed model performs about 3 ~ 4%p better than a method for predicting integrated labels. In addition, when learning Korean and English data using the proposed model, the performance was 1.7%p higher in English than in Korean due to the number of data and language disorder and the values of the parameters that produce the best performance were different. By excluding the number of directional cases, the proposed model can reduce the waste of resources in end-to-end relation extraction.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.10
• /
• pp.995-999
• /
• 2010

Every year, traffic accidents and traffic congestion have been rapidly increasing, Although the roadway design and signal system have been improved to relieve traffic accidents, traffic casualties and property damage do not decrease. This paper develops a real-time traffic accident detection and analysis system (RTADAS): In the proposed system, we aim to precisely detect traffic accidents at different design and flow of intersections, However, because the data collected from intersections have uncertainty and complicated causal dependency between them, we construct probability-based networks for correct accident detection.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.36S no.10
• /
• pp.103-112
• /
• 1999

Genetic algorithm, which uses principles of natural selection and population genetics, is an efficient method to find out an optimal solution. In conventional genetic algorithms, however, the size of gene pool needs to be increased to insure a convergency. Therefore, many memory spaces and much computation time were needed. Also, since child chromosomes were generated by chromosome crossover and gene mutation, the algorithms have a complex structure. Thus, in this paper, a compact stereo matching algorithm using a population-based incremental learning based on probability vector is proposed to reduce these problems. The PBIL method is modified for matching environment. Since th proposed algorithm uses a probability vector and eliminates gene pool, chromosome crossover, and gene mutation, the matching algorithm is simple and the computation load is considerably reduced. Even though the characteristics of images are changed, stable outputs are obtained without the modification of the matching algorithm.

• Journal of KIISE:Software and Applications
• /
• v.36 no.3
• /
• pp.208-216
• /
• 2009

A number of estimation of distribution algorithms have been proposed that do not use explicitly crossover and mutation of traditional genetic algorithms, but estimate the distribution of population for more efficient search. But because it is not easy to discover higher-order correlations of variables, lower-order correlations are estimated most cases under various constraints. In this paper, we propose a new estimation of distribution algorithm that represents higher-order correlations of the data and finds global optimum more efficiently. The proposed algorithm represents the higher-order correlations among variables by building random hypergraph model composed of hyperedges consisting of variables which are expected to be correlated, and generates the next population by Bayesian sampling algorithm Experimental results show that the proposed algorithm can find global optimum and outperforms the simple genetic algorithm and BOA(Bayesian Optimization Algorithm) on decomposable functions with deceptive building blocks.

• Korean Journal of Remote Sensing
• /
• v.27 no.1
• /
• pp.59-68
• /
• 2011

The analysis of optimal infiltration path is one of the representative fields in which the GIS technology can be useful for the military purpose. Usually the analysis of the optimal path is done with network data. However, for military purpose, it often needs to be done with raster data. Because raster data needs far more computation than network data, it is difficult to apply the methods usually used in network data, such as Dijkstra algorithm. The genetic algorithm, which has shown great outcomes in optimization problems, was applied. It was used to minimize the detection probability of infiltration route. 2D binary array genes and its crossover and mutation were suggested to solve this problem with raster data. 30 tests were performed for each population size, 500, 1000, 2000, and 3000. With each generation, more adoptable routes survived and made their children routes. Results indicate that as the generations increased, average detection probability decreased and the routes converged to the optimal path. Also, as the population size increases, more optimal routes were found. The suggested genetic algorithm successfully finds the optimal infiltration route, and it shows better performance with larger population.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2016.05a
• /
• pp.521-521
• /
• 2016

상수도 관망은 수원에서 수요절점까지 물을 안정적으로 공급하는 것을 목표로 한다. 상수도 관망의 최적설계는 수리학적 제한조건 (절점의 수압, 관로의 유속)을 만족하는 범위에서 비용을 최소화하는 설계안을 얻는 것으로 Savic and Walters (1997)는 유전 알고리즘 (Genetic Algorithms, Holland 1975)을 적용한 상수도 관망 설계 프로그램인 GANET를 제안하였고, Maier et al. (1996)은 개미군집알고리즘 (Ant Colony Optimization Algorithm, Dorigo et al. 1996)을 상수도 관망 최적설계에 적용한 후 그 결과가 유전 알고리즘에 비해 우수함을 증명하는 등 상수도 관망 최적설계에 관한 연구가 활발히 진행되어 오고 있다. 유전알고리즘은 선택, 교차, 돌연변이의 반복계산 과정을 통하여 최적해를 찾는 최적화 기법이다. 이 과정에서 결정변수는 유전자 (Gene)의 집합으로 표현되며, 염색체 (Chromosome) 내에서 근접한 유전 인자들은 일종의 Building Block을 형성하게 된다. Building Block은 좋은 해를 갖는 유전 인자를 높은 확률로 보관하여 지역해에 빠질 가능성을 줄이는 반면, 유전형 (Genotype)이 표현형 (Phenotype)을 충분히 모방하여 표현하지 못한 경우 오히려 최적해의 탐색을 방해할 수 있다는 한계점을 갖는다. 유전 알고리즘을 상수도 관망 최적설계에 적용하였을 때에도 이 한계점은 여실히 드러난다. 관로의 관경을 결정변수로 설정한 후 유전형으로 표현하였을 때, 관망도 상에서 근접하지 않은 두 관로가 염색체 내에서 연속으로 나열된다면 두 관로 간의 연관성이 실제보다 크게 고려되기 때문이다. 한편, 하모니써치 (Harmony Search, Geem et al. 2001) 알고리즘은 즉흥 연주 (Improvisation)를 통해 최상의 화음을 만들어내는 현상으로부터 착안하여 만들어진 최적화기법으로 연산 기법은 무작위선택, 기억회상, 피치조정 등으로 구성되어 있으며, 결정변수에 해당하는 연주자가 독립적으로 행동하며 해를 탐색한다는 점에서 유전알고리즘과 큰 차이를 갖는다. 본 연구에서는 유전알고리즘의 Building Block에 의해 발생하는 오류를 개선하고자, 상수도 관망 최적설계 연구에 많이 사용되는 Hanoi 관망 (Fujiwara and Khang 1990) 관로의 정렬 순서를 여러 가지 기준으로 설정하여 관망데이터를 구축한 후 하모니써치와 유전 알고리즘을 적용하여 최적화를 수행하였고 그 결과를 비교하였다. 그 결과 유전 알고리즘과 달리 하모니써치 알고리즘의 경우, 관로의 나열 순서와 상관없이 우수한 최적해 탐색 결과를 보이는 것을 확인할 수 있었다.

• Journal of KIISE:Software and Applications
• /
• v.28 no.7
• /
• pp.501-510
• /
• 2001

Generally, it is essential to estimate exact disparity for the 3D modeling from stereo images. Because existing methods calculate disparities from a whole image, they require too much cimputational time and bring about the mismatching problem. In this article, using the characteristic that the disparity vectors in stereo images are distributed not equally in a whole image but only exist about the background and obhect, we do a wavelet transformation on stereo images and estimate coarse disparity fields from the reduced lowpass field using area-based method at first-step. From these coarse disparity vectors, we generate disparity histogram and then separate object from background area using it. Afterwards, we restore only object area to the original image and estimate dense and accurate disparity by our two-step pixel-based method which does not use pixel brightness but use second gradient. We also extract feature points from the separated object area and estimate depth information by applying disparity vectors and camera parameters. Finally, we generate 3D model using both feature points and their z coordinates. By using our proposed, we can considerably reduce the computation time and estimate the precise disparity through the additional pixel-based method using LOG filter. Furthermore, our proposed foreground/background method can solve the mismatching problem of existing Delaunay triangulation and generate accurate 3D model.