• 제어로봇시스템학회논문지
• /
• 제2권3호
• /
• pp.165-173
• /
• 1996

It is not desirable to have too large control input in control systems, because there are usually a limitation for the input magnitude and cost for the input energy. Previous papers in the iterative learning control did not considered on these points. In this paper, an iterative learning control method is proposed for a class of nonlinear systems with consideration on input magnitude by adopting a concept of cost function consisting of the output error and the input magnitude in quadratic form. We proposed a new input update law with an input penalty function. If we choose a reasonable input penalty function, the two control objectives, good command following and small input energy, can be achieved. The characteristics of the proposed method are shown in the simulation examples.

• 한국멀티미디어학회논문지
• /
• 제24권1호
• /
• pp.13-20
• /
• 2021

A Wasserstein GAN(WGAN), optimum in terms of minimizing Wasserstein distance, still suffers from inconsistent convergence or unexpected output due to inherent learning instability. It is widely known some kinds of restriction on the discriminative function should be considered to solve such problems, which implies the importance of Lipschitz continuity. Unfortunately, there are few known methods to satisfactorily maintain the Lipschitz continuity of the discriminative function. In this paper we propose techniques to stably maintain the Lipschitz continuity of the discriminative function by adding effective regularization terms to the objective function, which limit the magnitude of the gradient vectors of the discriminator to one or less. Extensive experiments are conducted to evaluate the performance of the proposed techniques, which shows the single-sided penalty improves convergence compared with the gradient penalty at the early learning process, while the proposed additional penalty increases inception scores by 0.18 after 100,000 number of learning.

• 로봇학회논문지
• /
• 제17권1호
• /
• pp.40-47
• /
• 2022

A reward function suitable for a task is required to manipulate objects through reinforcement learning. However, it is difficult to design the reward function if the ample information of the objects cannot be obtained. In this study, a demonstration-based object manipulation algorithm called stochastic exploration guided by demonstration (SEGD) is proposed to solve the design problem of the reward function. SEGD is a reinforcement learning algorithm in which a sparse reward explorer (SRE) and an interpolated policy using demonstration (IPD) are added to soft actor-critic (SAC). SRE ensures the training of the critic of SAC by collecting prior data and IPD limits the exploration space by making SEGD's action similar to the expert's action. Through these two algorithms, the SEGD can learn only with the sparse reward of the task without designing the reward function. In order to verify the SEGD, experiments were conducted for three tasks. SEGD showed its effectiveness by showing success rates of more than 96.5% in these experiments.

• 반도체디스플레이기술학회지
• /
• 제22권2호
• /
• pp.40-44
• /
• 2023

The defect detection based on deep learning requires minimal loss and high accuracy to pinpoint product defects. In this paper, we confirm the loss rate of deep learning training based on disc-shaped artificial scaffold images. It is intended to compare the performance of Cross-Entropy functions used in object detection algorithms. The model was constructed using normal, defective artificial scaffold images and category cross entropy and sparse category cross entropy. The data was repeatedly learned five times using each loss function. The average loss rate, average accuracy, final loss rate, and final accuracy according to the loss function were confirmed.

• 산업공학
• /
• 제19권4호
• /
• pp.324-332
• /
• 2006

It is believed that each learner has a preferred method to acquire and manage knowledge according to her/his learning style which influences learning achievement directly. The purpose of this paper is to statistically analyze relationships among individual learning styles, tutoring functions, and learning achievement in an e-learning environment. 524 survey results from participants of enterprise e-learning classes are classified into total group and superior group. T-Test and ANOVA analyses are carried between learning style and learning achievement and between learning style and preferred tutoring functions. The analysis results show that individual learning styles do not contribute to learning achievement while they are strongly related to preferences for some of tutoring functions. These results can be used to identify limitation of current e-learning practice and design better e-learning systems, especially, supporting appropriate tutoring functions for different types of learners.

• 전기전자학회논문지
• /
• 제23권4호
• /
• pp.1187-1194
• /
• 2019

본 논문에서는 인공지능 오델로 게임 에이전트를 구현하기 위해 실제 프로기사들의 기보를 CNN으로 학습시키고 이를 상태의 형세 판단을 위한 근거로 삼아 최소최대탐색을 이용해 현 상태에서 최적의 수를 찾는 의사결정구조를 사용하고 이를 발전시키고자 강화학습 이론을 이용한 자가대국 학습방법을 제안하여 적용하였다. 본 논문에서 제안하는 구현 방법은 기보학습의 성능 평가 차원에서 가치평가를 위한 네트워크로서 기존의 ANN을 사용한 방법과 대국을 통한 방법으로 비교하였으며, 대국 결과 흑일 때 69.7%, 백일 때 72.1%의 승률을 나타내었다. 또한 본 논문에서 제안하는 강화학습 적용 결과 네크워크의 성능을 강화학습을 적용하지 않은 ANN 및 CNN 가치평가 네트워크 기반 에이전트와 비교한 결과 각각 100%, 78% 승률을 나타내어 성능이 개선됨을 확인할 수 있었다.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1998년도 추계학술대회 논문집 학회본부 B
• /
• pp.407-409
• /
• 1998

This paper deals with temporal-difference learning that is a method for approximating long-term future cost as a function of current state in knowlege-poor environment, a function approximator is used to approximate the mapping from state to future cost, a linear function approximator is limited because mapping from state to future cost has a nonlinear characteristic, so a nonlinear function approximator is used to approximate the mapping from state to future cost in this paper, and that TD learning using a nonlinear function approximator is stable is proved.

• 한국콘텐츠학회논문지
• /
• 제11권12호
• /
• pp.969-975
• /
• 2011

본 논문은 대학 이러닝 수업에서 학생들의 자발적인 학습활동과 인터넷 수업기능의 활용, 학습 성공에 대한 이해정도가 학습 성취도에 미치는 영향을 분석하였다. 대학에서 이러닝 수업을 듣고 있는 학생들 297명을 대상으로 설문분석을 통하여 핵심변인에 대한 요인들을 분석한 결과 인터넷 수업 시 자기주도성을 나타내는 다양한 요인들이 학생들의 자발성(자기주도성), 인터넷수업기능 활용성, 학습에 대한 성공요인 이해정도의 세 가지 카테고리로 분류되었다. 이 세 가지 요인변수들이 결과변수인 학생들의 성적에 미치는 영향을 분석한 연구결과, 자기주도적인 성향이 학습자 성취도에 유의미한 영향이 있는 것으로 나타났다. 이를 통해 대학 이러닝 수업에서 다양한 상호작용적인 인터넷 활용기능이나 학습 성공에 대한 이해정도보다 학생들의 자발적 수업활동에 대한 직접적인 준비가 학습 성취에 결정적인 영향을 미칠 수 있다는 것을 알 수 있었다. 이는 향후 대학의 이러닝 수업에서 교수 학습 전략을 설계할 때 고려해야 하는 중요한 시사점이라 할 수 있다.

• 무역학회지
• /
• 제46권6호
• /
• pp.191-209
• /
• 2021

In this paper, various time series analysis models and machine learning models are presented for long-term prediction of export growth rate, and the prediction performance is compared and reviewed by RMSE and MAE. Export growth rate is one of the major economic indicators to evaluate the economic status. And It is also used to predict economic forecast. The export growth rate may have a negative (-) value as well as a positive (+) value. Therefore, Instead of using the ReLU function, which is often used for time series prediction of deep learning models, the PReLU function, which can have a negative (-) value as an output value, was used as the activation function of deep learning models. The time series prediction performance of each model for three types of data was compared and reviewed. The forecast data of long-term prediction of export growth rate was deduced by three forecast methods such as a fixed forecast method, a recursive forecast method and a rolling forecast method. As a result of the forecast, the traditional time series analysis model, ARDL, showed excellent performance, but as the time period of learning data increases, the performance of machine learning models including LSTM was relatively improved.

• 전자공학회논문지B
• /
• 제32B권3호
• /
• pp.503-511
• /
• 1995

A batch-mode algorithm is proposed to increase the speed of learning in the error backpropagation algorithm with variable learning rate and variable momentum parameters in classification problems. The objective function is normalized with respect to the number of patterns and output nodes. Also the gradient of the objective function is normalized in updating the connection weights to increase the effect of its backpropagated error. The learning rate and momentum parameters are determined from a function of the gradient norm and the number of weights. The learning rate depends on the square rott of the gradient norm while the momentum parameters depend on the gradient norm. In the two typical classification problems, simulation results demonstrate the effectiveness of the proposed algorithm.