• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.3
• /
• pp.15-20
• /
• 2024

Path planning for unmanned aerial vehicles (UAV) is crucial in avoiding collisions with obstacles in complex environments that include both static and dynamic obstacles. Path planning algorithms like RRT and A^* are effectively handle static obstacle avoidance but have limitations with increasing computational complexity in high-dimensional environments. Reinforcement learning-based algorithms can accommodate complex environments, but like traditional path planning algorithms, they struggle with training complexity and convergence in higher-dimensional environment. In this paper, we proposed a reinforcement learning model utilizing a cell decomposition algorithm. The proposed model reduces the complexity of the environment by decomposing the learning environment in detail, and improves the obstacle avoidance performance by establishing the valid action of the agent. This solves the exploration problem of reinforcement learning and improves the convergence of learning. Simulation results show that the proposed model improves learning speed and efficient path planning compared to reinforcement learning models in general environments.

• Proceedings of the KIEE Conference
• /
• 2001.11b
• /
• pp.317-319
• /
• 2001

송전선에 사고가 발생하면 계전기가 동작하고 여러 가지 후속 조치가 적절히 수행되기 위해서는 빠른 시간 내에 고장의 검출, 고장 종류의 분류, 고장 위치의 판정 등이 요구된다. 신경회로망을 이용한 고장 해석은 빠르고 정확한 해석이 가능하기 때문에 최근에 각광을 받고 있다. 본 논문에서는 기존의 코호넨 신경회로망의 지도 경쟁 학습 알고리즘을 개선한 새로운 학습 알고리즘이 제시된다. 제안된 학습 알고리즘은 전력계통의 송전선 고장 종류 분류에 적용된다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2002.04b
• /
• pp.454-456
• /
• 2002

한국어의 비재귀 명사구 즉 기반 명사구(basehp)를 인식하는 알고리즘을 제시한다. 본 논문에서는 한개의 주어진 학습 알고리즘에 대해 문맥 윈도우의 크기와 문맥 윈도우의 위치를 달리해 가면서 학습시킨다 이러한 방법을 통해 서로 다른 정보를 바탕으로 한 기반 명사구 인식을 수행할 수 있으며, 그 결과서로 다른 여러 개의 결과들을 생성할 수 있다. 본 논문에에서는 이렇게 얻어진 여러 개의 인식 결과들을 적절한 방법으로 결합하여 한국어에서 91% 이상의 높은 기반명사구 인식 정확도를 얻어낼 수 있다. 15만 단어 규모의 국어정보베이스의 말뭉치를 사용했으며 , 학습 알고리즘으로는 메모리 기반 학습 알고리즘 (memory-based learning)을 이용하여 실험하였다.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.3
• /
• pp.747-753
• /
• 1997

Function approximation from a set of input-output parirs has numerous applications in scientiffc and engineer-ing areas.Multiayer feedforward neural networks have been proposed as a good approximator of noninear function.The back propagation (BP) algorithm allows muktiayer feedforward neural networks oro learn input-output mappongs from training samples.However, the mapping acquired through the BP algorithm nay be cor-rupt when errorneous trauning data are employed.In this paper we propose a robust BP learning algorithm that is resistant to the errormeous data and is capable of rejecting gross errors during the approximation process.

• The Transactions of the Korea Information Processing Society
• /
• v.5 no.11
• /
• pp.2897-2907
• /
• 1998

The production-rule generation from training examples is a hard problem that has large space and many local optimal solutions. Many learning methods are proposed for production-rule generation and genetic algorithms is an alternative learning method. However, traditional genetic algorithms has been known to have an obstacle in converging at the global solution area and show poor efficiency of production-rules generated. In this paper, we propose a production-rule generating method which uses genetic algorithm learning. By analyzing optimal sub-solutions captured by genetic algorithm learning, our method takes advantage of its schema structure and thus generates relatively small rule set.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.41 no.2
• /
• pp.87-94
• /
• 2004

The feature subset selection is used as a preprocessing step of a teaming algorithm. If collected data are irrelevant or redundant information, we can improve the performance of learning by removing these data before creating of the learning model. The feature subset selection can also reduce the search space and the storage requirement. This paper proposed a new feature subset selection algorithm that is using the heuristic function based on entropy to evaluate the performance of the abstracted feature subset and feature selection. The ACS algorithm was used as a search method. We could decrease a size of learning model and unnecessary calculating time by reducing the dimension of the feature that was used for learning.

• Journal of Korea Society of Industrial Information Systems
• /
• v.9 no.4
• /
• pp.47-55
• /
• 2004

Most questions made for remote examinations on web-based education system use methods of making questions using fixed questions or randomly using item pools or automatically using degree of difficulty. Particularly, automatically selection methods using degree of difficulty is the kernel of a question that objectivity of examination questions by degree of difficulty adjustment based result of examination. This paper is use automatically selection methods for examination on web-based education system. Therefore we present new algorithms of mediateness degree of difficulty as regards education ability of students for adjust the degree of difficulty. We identified this algorithms is more effective as compared with previously algorithms on web-based education system

• Journal of IKEEE
• /
• v.17 no.3
• /
• pp.254-259
• /
• 2013

The choice of labeled data in semi-supervised learning algorithm can result in effects on the performance of the resultant classifier. In order to select labeled data required for the training of a semi-supervised learning algorithm, VCNN(Vector Centroid Neural Network) is proposed in this paper. The proposed selection method of label data is evaluated on UCI dataset and caltech dataset. Experiments and results show that the proposed selection method outperforms conventional methods in terms of classification accuracy and minimum error rate.

• The KIPS Transactions:PartB
• /
• v.14B no.5
• /
• pp.377-382
• /
• 2007

This paper proposes a training algorithm for global optimization of the parameters of radial basis function networks. Since conventional training algorithms usually perform only local optimization, the performance of the network is limited and the final network significantly depends on the initial network parameters. The proposed hybrid simulated annealing algorithm performs global optimization of the network parameters by combining global search capability of simulated annealing and local optimization capability of gradient-based algorithms. Via experiments for function approximation problems, we demonstrate that the proposed algorithm can find networks showing better training and test performance and reduce effects of the initial network parameters on the final results.

• Proceedings of the KIEE Conference
• /
• 2007.04a
• /
• pp.307-309
• /
• 2007

SVDD의 scale prob1em을 해결하기 위하여, 학습 데이터를 sub-groupings하여 group 단위로 SVDD를 통해 학습함으로써 학습 시간을 줄이는, K-means clustering을 이용한 SVDD 방범(KMSVDD)이 제안되었다. 하지만 KMSVDD는 K-means clustering 알고리즘의 본질상 최적의 K값을 정하기 힘들다는 문제와, 동일한 데이터를 학습할지라도 clustered group이 램덤하게 형성되기 때문에 매번 학습의 결과가 달라지는 문제점이 있었다. 또한 데이터의 분포 상태와 관계없이 무조건 타원(dlliptic) 형태의 K개의 cluster로 나누기 때문에 각각의 나눠진 cluster들은 데이터 분포에 대한 특징을 나타내기 힘들게 된다. 이러한 문제점을 해결하기 위하여 본 논문에서는 데이터 분포에서 mode를 먼저 찾은 후 이 mode를 기준으로 clustering하는 Mean Shift clustering 방법을 이용한 SVDD를 제안하고자 한다. 제안된 알고리즘은 KMSVDD와 비교해 데이터 학습 속도에서는 큰 차이가 없으면서도 데이터의 분포 상태를 고려한 형태로 clustering 한 sub-group을 학습하므로 학습의 정확도가 일정하게 되며, 각각의 cluster는 데이터 분표의 특징을 포함하는 효과가 있다. 또한 Mean Shift Kernel의 bandwidth의 결정은 K-Means의 K와는 달리 어느 정도 여유를 갖고 결정되어도 학습 결과에는 차이가 없다. 다양한 데이터들을 이용한 모의실험을 통하여 위의 내용들을 검증하도록 한다.