• Proceedings of the KIEE Conference
• /
• 1998.07b
• /
• pp.740-743
• /
• 1998

This paper presents an optimization algorithm for a stable Dynamic Neural Network (DNN) using genetic algorithm. Optimized DNN is applied to a problem of controlling nonlinear dynamical systems. DNN is dynamic mapping and is better suited for dynamical systems than static forward neural network. The real time implementation is very important, and thus the neuro controller also needs to be designed such that it converges with a relatively small number of training cycles. SDNN has considerably fewer weights than DNN. The object of proposed algorithm is to the number of self dynamic neuron node and the gradient of activation functions are simultaneously optimized by genetic algorithms. To guarantee convergence, an analytic method based on the Lyapunov function is used to find a stable learning for the SDNN. The ability and effectiveness of identifying and controlling, a nonlinear dynamic system using the proposed optimized SDNN considering stability' is demonstrated by case studies.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2021.01a
• /
• pp.257-260
• /
• 2021

본 논문에서는 카메라의 포커싱과 아웃포커싱에 의해 이미지에서 뿌옇게 표현되는 DoF(Depth of field, 피사계 심도) 영역을 쿼드트리(Quadtree) 기반의 합성곱 신경망을 통해 빠르게 찾는 방법을 제안한다. 우리의 접근 방식은 RGB채널기반의 상호-상관 필터를 이용하여 DoF영역을 이미지로부터 효율적으로 분류하고, 적응형 트리인 쿼드트리를 기반으로 유의미한 영역만을 분류한다. 이 과정에서 손실 없이 온전하게 DoF영역을 추출하기 위한 필터링 과정을 거친다. 이러한 과정에서 얻어진 이미지 패치들은 전체 이미지에 비해 적은 영역으로 나타나며, 이 적은 개수의 패치들을 이용하여 네트워크 단계에서 사용할 이미지-DoF가중치 맵 데이터 쌍을 설정한다. 네트워크 과정에서 학습할 때 사용되는 데이터는 이미지와 상호-상관 필터 기반으로 추출된 DoF 가중치 맵을 이용한다. 본 논문에서 제안하는 쿼드트리 기반 합성곱 신경망은 이미지로부터 포커싱과 아웃포커싱된 DoF영역을 자동으로 추출하는 과정을 학습시키기 위해 사용된다. 결과적으로 학습에 필요한 데이터 영역이 줄어듦으로써 학습 시간과 메모리를 절약했으며, 테스트 결과로 얻은 DoF 가중치 이미지는 입력 이미지에서 DoF영역을 더욱더 빠른 시간 내에 찾아낸다.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.60 no.1
• /
• pp.87-99
• /
• 2024

This study deals with the application of an artificial neural network (ANN) model to predict power consumption for utilizing seawater source heat pumps of recirculating aquaculture system. An integrated dynamic simulation model was constructed using the TRNSYS program to obtain input and output data for the ANN model to predict the power consumption of the recirculating aquaculture system with a heat pump system. Data obtained from the TRNSYS program were analyzed using linear regression, and converted into optimal data necessary for the ANN model through normalization. To optimize the ANN-based power consumption prediction model, the hyper parameters of ANN were determined using the Bayesian optimization. ANN simulation results showed that ANN models with optimized hyper parameters exhibited acceptably high predictive accuracy conforming to ASHRAE standards.

• Proceedings of the IEEK Conference
• /
• 1998.10a
• /
• pp.1281-1284
• /
• 1998

This paper proposes an efficient time series prediction of the nonlinear dynamical discrete-time systems using multilayer neural networks of a hybrid learning algorithm. The proposed learning algorithm is a hybrid backpropagation algorithm based on the steepest descent for high-speed optimization and the dynamic tunneling for global optimization. The proposed algorithm has been applied to the y00 samples of 700 sequences to predict the next 100 samples. The simulation results shows that the proposed algorithm has better performances of the convergence and the prediction, in comparision with that using backpropagation algorithm based on the gradient descent for multilayer neural network.

• Proceedings of the Korean Institute of IIIuminating and Electrical Installation Engineers Conference
• /
• 2007.05a
• /
• pp.322-327
• /
• 2007

This paper is proposed an efficiency optimization control algorithm for IPMSM which minimizes the copper and iron losses. The design of the speed controller based on adaptive fuzzy learning control-fuzzy neural networks(ABLC-FNN) controller that is implemented using adaptive, fuzzy control and neural networks. The control performance of the AFLC-FNN controller is evaluated by analysis for various operating conditions. Analysis results are presented to show the validity of the proposed algorithm

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.39 no.4
• /
• pp.403-413
• /
• 1990

A collision free trajectory control for multiple mobile robots in obstacle-resident workspace is proposed. The proposed method is based on the concept of neural optimization network which has been applied to such problems which are too complex to be handled by traditional analytical methods, and gives good adaptibility for unpredictable environment. In this paper, the positions of the mobile robot are taken as the variables of the neural circuit and the differential equations are derived based on the performance index which is the weighted summation of the functions of the distances between the goal and current position of each robot, between each pair of robots and between the goal and current position of each robot, between each pair of robots and between obstacles and robots. Also is studied the problem of local minimum and of detour in large radius around obstacles, which is caused by inertia of mobile robots. To show the validity of the proposed method an example is illustrated by computer simulation, in which 6 mobile robots with mass and friction traverse in a workspace with 6 obstacles.

• Bulletin of the Korean Chemical Society
• /
• v.26 no.3
• /
• pp.399-408
• /
• 2005

We have utilized neural network (NN) studies to predict impact sensitivities of various types of explosive molecules. Two hundreds and thirty four explosive molecules have been taken from a single database, and thirty nine molecular descriptors were computed for each explosive molecule. Optimization of NN architecture has been carried out by examining seven different sets of molecular descriptors and varying the number of hidden neurons. For the optimized NN architecture, we have utilized 17 molecular descriptors which were composed of compositional and topological descriptors in an input layer, and 2 hidden neurons in a hidden layer.

• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.48 no.2
• /
• pp.82-88
• /
• 1999

In this paper, we propose a new structure optimization method of multilayer neural networks which begin and carry out learning from a bigger network. This method redundant links and neurons according to the rough set theory. In order to find redundant links, we analyze the variations of all weights and output errors in every step of the learning process, and then make the decision table from their variation of weights and output errors. We can find the redundant links from the initial structure by analyzing the decision table using the rough set theory. This enables us to build a structure as compact as possible, and also enables mapping between input and output. We show the validity and effectiveness of the proposed algorithm by applying it to the XOR problem.

• Electronics and Telecommunications Trends
• /
• v.34 no.4
• /
• pp.32-42
• /
• 2019

The performance of machine learning algorithms significantly depends on how a configuration of hyperparameters is identified and how a neural network architecture is designed. However, this requires expert knowledge of relevant task domains and a prohibitive computation time. To optimize these two processes using minimal effort, many studies have investigated automated machine learning in recent years. This paper reviews the conventional random, grid, and Bayesian methods for hyperparameter optimization (HPO) and addresses its recent approaches, which speeds up the identification of the best set of hyperparameters. We further investigate existing neural architecture search (NAS) techniques based on evolutionary algorithms, reinforcement learning, and gradient derivatives and analyze their theoretical characteristics and performance results. Moreover, future research directions and challenges in HPO and NAS are described.

• Korean Journal of Artificial Intelligence
• /
• v.12 no.3
• /
• pp.33-39
• /
• 2024

In this paper, we explore the application of Kodaly hand signs in enhancing children's music education, performances, and auditory assistance technologies. This research focuses on improving the recognition rate of Multilayer Perceptron (MLP) models in identifying Kodaly hand sign scales through the integration of Artificial Neural Networks (ANN). We developed an enhanced MLP model by augmenting it with additional parameters and optimizing the number of hidden layers, aiming to substantially increase the model's accuracy and efficiency. The augmented model demonstrated a significant improvement in recognizing complex hand sign sequences, achieving a higher accuracy compared to previous methods. These advancements suggest that our approach can greatly benefit music education and the development of auditory assistance technologies by providing more reliable and precise recognition of Kodaly hand signs. This study confirms the potential of parameter augmentation and hidden layers optimization in refining the capabilities of neural network models for practical applications.