• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.174-174
• /
• 2000

Q-learning, based on discrete state and action space, is a most widely used reinforcement Learning. However, this requires a lot of memory and much time for learning all actions of each state when it is applied to a real mobile robot navigation using continuous state and action space Region-based Q-learning is a reinforcement learning method that estimates action values of real state by using triangular-type action distribution model and relationship with its neighboring state which was defined and learned before. This paper proposes a new Region-based Q-learning which uses a reward assigned only when the agent reached the target, and get out of the Local optimal path with adjustment of random action rate. If this is applied to mobile robot navigation, less memory can be used and robot can move smoothly, and optimal solution can be learned fast. To show the validity of our method, computer simulations are illusrated.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2013.10a
• /
• pp.664-665
• /
• 2013

This paper present a method of constructing the multimedia processor architecture. The proposed multimedia processor architecture be able to handle each text, sound, and video in one chip. Also it have interactive function that is a characteristics of multimedia. Specially, the proposed multimedia processor be able to addressing nodes in memory map without software, and it is completely reconfigurable depend on data. Also it as able to process time and space common that have synchronous/asynchronous and it is able to protect continuous and dynamic media bus collision, and local and overall common memory structure. The proposed multimedia processor architecture apply to virtual reality and mixed reality.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.13 no.4
• /
• pp.277-283
• /
• 2013

In this study, we propose a novel method for modeling dynamic hysteresis. Hysteresis is a widespread phenomenon that is observed in many physical systems. Many different models have been developed for representing a hysteretic system. Among them, the Duhem model is a classical nonlinear dynamic hysteresis model satisfying the properties of hysteresis. The purpose of this work is to develop a novel method that expresses the local dynamics of the Duhem model by a linear system model. Our approach utilizes a certain type of fuzzy system that is based on Takagi-Sugeno (T-S) fuzzy models. The proposed T-S fuzzy Duhem model is achieved by fuzzy blending of the linear system model. A simulated example applied to shape memory alloy actuators, which have typical hysteretic properties, illustrates the applicability of our proposed scheme.

• Journal of the Semiconductor & Display Technology
• /
• v.19 no.2
• /
• pp.82-87
• /
• 2020

In this paper, we introduce the novel idea to improve the B/W usage efficiency of on-chip networks used for TAM to test multiple DRAMs. In order to avoid the local bottleneck of test packets caused by an ATE, we make test patterns using microcode-based instructions within ATE and adopt a test bus to transmit test responses from DRAM DFT (Design for Testability) called Test Generator (TG) to ATE. The proposed test platform will contribute to increasing the test economics of memory IC industry.

• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 2004.07a
• /
• pp.37-40
• /
• 2004

SONOS를 이용한 전하트랩형 플래시 메모리를 통상의 0.35um CMOS 공정을 이용하여 제작하였으며 그 구조는 소스를 공통(CSL. Common Source Line)으로 사용하는 NOR형으로 하였다. 기존의 공정을 그대로 이용하면서 멀티 비트 동작을 통한 실질적 집적도 향상을 얻을 수 있다면 그 의미가 크다고 하겠다. 따라서 본 연구에서는CSL-NOR형 플래시 구조에서 멀티 비트을 구현하기위한 최적의 프로그램/소거/읽기 전압 조건을 구하여 국소적으로 트랩된 전하의 분포를 전하펌핑 방법을 이용하여 조사하였다. 또한 이 방법을 이용하여 멀티 비트 동작 시 문제점으로 제시된 전하의 측면확산을 측정하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.98-101
• /
• 2019

At present, deep convolutional neural networks have made a very important contribution in single-image super-resolution. Through the learning of the neural networks, the features of input images are transformed and combined to establish a nonlinear mapping of low-resolution images to high-resolution images. Some previous methods are difficult to train and take up a lot of memory. In this paper, we proposed a simple and compact deep recursive residual network learning the features for single image super resolution. Global residual learning and local residual learning are used to reduce the problems of training deep neural networks. And the recursive structure controls the number of parameters to save memory. Experimental results show that the proposed method improved image qualities that occur in previous methods.

• Journal of Korea Multimedia Society
• /
• v.23 no.8
• /
• pp.1019-1029
• /
• 2020

In this paper, we propose a novel knowledge distillation algorithm to create an compressed deep ensemble network coupled with the combined use of local and global features of face images. In order to transfer the capability of high-level recognition performances of the ensemble deep networks to a single deep network, the probability for class prediction, which is the softmax output of the ensemble network, is used as soft target for training a single deep network. By applying the knowledge distillation algorithm, the local feature informations obtained by training the deep ensemble network using facial subregions of the face image as input are transmitted to a single deep network to create a so-called compressed ensemble DCNN. The experimental results demonstrate that our proposed compressed ensemble deep network can maintain the recognition performance of the complex ensemble deep networks and is superior to the recognition performance of a single deep network. In addition, our proposed method can significantly reduce the storage(memory) space and execution time, compared to the conventional ensemble deep networks developed for face recognition.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.11
• /
• pp.151-162
• /
• 1994

A new neural network structure called Self-organized Distributed Networks (SODN) is proposed for developing the neural network-based multidimensional system models. The learning with the proposed networks is fast and precise. Such properties are caused from the local learning mechanism. The structure of the networks is combination of dual networks such as self-organized networks and multilayered local networks. Each local networks learns only data in a sub-region. Large number of memory requirements and low generalization capability for the untrained region, which are drawbacks of conventional local network learning, are overcomed in the proposed networks. The simulation results of the proposed networks show better performance than the standard multilayer neural networks and the Radial Basis function(RBF) networks.

• Journal of Institute of Control, Robotics and Systems
• /
• v.12 no.7
• /
• pp.693-699
• /
• 2006

An accurate and compact map is essential to an autonomous mobile robot system. For navigation, it is efficient to use an occupancy grid map because the environment is represented by probability distribution. But it is difficult to apply it to the large environment since it needs a large amount of memory proportional to the environment size. As an alternative, a topological map can be used to represent it in terms of the discrete nodes with edges connecting them. It is usually constructed by the Voronoi-like graphs, but in this paper the topological map is incrementally built based on the local grid map using the thinning algorithm. This algorithm can extract only meaningful topological information by using the C-obstacle concept in real-time and is robust to the environment change, because its underlying local grid map is constructed based on the Bayesian update formula. In this paper, the position probability is defined to evaluate the quantitative reliability of the end nodes of this thinning-based topological map (TTM). The global TTM can be constructed by merging each local TTM by matching the reliable end nodes determined by the position probability. It is shown that the proposed TTM can represent the environment accurately in real-time and it is readily extended to the global TTM.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.9
• /
• pp.4412-4428
• /
• 2018

The purpose of this work is to solve the problem of representing an entire video using Convolutional Neural Network (CNN) features for human action recognition. Recently, due to insufficient GPU memory, it has been difficult to take the whole video as the input of the CNN for end-to-end learning. A typical method is to use sampled video frames as inputs and corresponding labels as supervision. One major issue of this popular approach is that the local samples may not contain the information indicated by the global labels and sufficient motion information. To address this issue, we propose a binary hashing method to enhance the local feature extractors. First, we extract the local features and aggregate them into global features using maximum/minimum pooling. Second, we use the binary hashing method to capture the motion features. Finally, we concatenate the hashing features with global features using different normalization methods to train the classifier. Experimental results on the JHMDB and MPII-Cooking datasets show that, for these new local features, binary hashing mapping on the sparsely sampled features led to significant performance improvements.