http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Kim, Hyunah,Kim, Chayoung The International Promotion Agency of Culture Tech 2021 International Journal of Advanced Culture Technolo Vol.9 No.1
Pervasive enhancement and required enforcement of the Internet of Things (IoTs) in a distributed massively multiplayer online architecture have effected in massive growth of Big-Data in terms of server over-load. There have been some previous works to overcome the overloading of server works. However, there are lack of considered methods, which is commonly applicable. Therefore, we propose a combing Sparse Auto-Encoder and Load-Balancing, which is ProGReGA for Big-Data of server loads. In the process of Sparse Auto-Encoder, when it comes to selection of the feature-pattern, the less relevant feature-pattern could be eliminated from Big-Data. In relation to Load-Balancing, the alleviated degradation of ProGReGA can take advantage of the less redundant feature-pattern. That means the most relevant of Big-Data representation can work. In the performance evaluation, we can find that the proposed method have become more approachable and stable.
Goal-based Target Network in Deep Q-Network with Hindsight Experience Replay
Chayoung Kim 한국정보기술학회 2021 한국정보기술학회논문지 Vol.19 No.7
Handling a sparse reward is one of the most significant challenges in Reinforcement Learning(RL), especially when we achieve human-level performances in a complex domain, such as grasping a moving object of robotic manipulation or preventing a corrupted byte in a cipher text. Deep Q-Network(DQN) with Hindsight Experience Replay(HER) is effective in dealing with sparse rewards, but it is not easy to get the advantages of on-line learning using target network, which has a fixed update. Therefore, in relation to sparse compensation, we propose a method of updating the target network based on the goal of HER in DQN. Since the Goal-based Target Network for HER in DQN proposed in this paper is updated every episode and every goal, it is more frequent and flexible so that the advantages of on-line learning can be utilized a little more. We evaluate the proposed Goal-based Target Network on a bit-flipping environment for preventing Byte-Flipping-Attack. The comparison demonstrates the superiority of our approach, showing that the proposed Goal-based Target Network is a better ingredient to enable the HER in DQN to solve tasks on the domain of the sparse reward.
Scheme of a Classic Control-Based Program Model with Non-Symmetric Deep Auto-Encoder of Actor-Critic
Chayoung Kim 한국정보기술학회 2020 한국정보기술학회논문지 Vol.18 No.7
Reinforcement learning (RL), particularly Actor-Critic (A2C), one of policy gradient (PG) algorithms becomes a mainstream. A design of a classic control-based program model requires domain-indicated experience knowledge based on a neural network for incorporating into the model strategy, which becomes A2C-powered. There are some research studies on the program model using general Artificial Intelligence (AI) techniques, in other words, A2C and neural networks. The previous well-known algorithms can proceed from a simple Convolutional Neural Network (CNN), to attempting an experiment with more complicated additions, such as LSTM (Long Short-Term Memory models). However, there are concerns about the requirements of the current experimental environment when faced with the demands of significant computational powers. Therefore, we employ a definition of Actor-Critic with non-symmetric deep auto-encoder to find an optimal behavior strategy for the agent to obtain optimal rewards with the bottommost environmental resources. Algorithm comparisons show that our proposed model demonstrates improvements of optimal rewards with the minimum limit over the developed deep neural network combined Actor-Critic.
A MA-plot-based Feature Selection by MRMR in SVM-RFE in RNA-Sequencing Data
Chayoung Kim 한국정보기술학회 2018 한국정보기술학회논문지 Vol.16 No.12
It is extremely lacking and urgently required that the method of constructing the Gene Regulatory Network (GRN) from RNA-Sequencing data (RNA-Seq) because of Big-Data and GRN in Big-Data has obtained substantial observation as the interactions among relevant featured genes and their regulations. We propose newly the computational comparative feature patterns selection method by implementing a minimum-redundancy maximum-relevancy (MRMR) filter the support vector machine-recursive feature elimination (SVM-RFE) with Intensity-dependent normalization (DEGSEQ) as a preprocessor for emphasizing equal preciseness in RNA-seq in Big-Data. We found out the proposed algorithm might be more scalable and convenient because of all libraries in R package and be more improved in terms of the time consuming in Big-Data and minimum-redundancy maximum-relevancy of a set of feature patterns at the same time.