Bae-Ho Lee

An effective machine learning algorithm using momentum scheduling

This paper proposes a new algorithm to improve learning performance in support vector machine by using the kernel relaxation and the dynamic momentum. Compared with the static momentum, the dynamic momentum is simultaneously obtained by the learning process of pattern weight and reflected into different momentum by the current state. Therefore, the proposed dynamic momentum algorithm can effectively control the convergence rate and performance. The experiment using SONAR data shows that the proposed algorithm has better convergence rate and performance than the kernel relaxation using static momentum.

Load-Sharing Algorithm using Digital Parallel Communication

In this paper, we proposes a new load-sharing algorithm with a ATmega2560 based digital communication. Proposed algorithm is different from conventional analog method. The high speed communication digital control is performed. To apply the digital communication and real-time control for time-sharing token bus method, we implemented high efficient load-sharing and redundancy. Also this system make down the price by auto ID algorithm and system response is improved by controller`s voltage and current integral value sharing. In parallel system prototype, each module have controller and performed load-sharing according to master module integral value. In this paper, we verify the validity of proposed algorithm using PSIM program and prototype.

A Study of Parallel Operation of Module Power using CAN Communication

In this paper, we proposes new load-sharing algorism for equal current division using CAN communication. Proposed algorithm is different from conventional analog method, it performed strong Load-sharing using bi-direction high speed communication. Each modules constitution on independence controller (voltage controller, electric current controller). In parallel system prototype, each module have controller and performed load-sharing according to master module integral value. Also additional controller use for getting each module situations that fault situation of module and fault locate of module . we implemented high efficient load-sharing and redundancy. In this paper, we verify the validity of proposed algorithm using PSIM program and prototype.

An efficient Bi-directional flooding in wireless sensor networks

Broadcast has been widely used in wireless sensor networks to disseminate information to all reachable nodes. The simplest way of realizing broadcast is via omnidirectional flooding. However, it generates an excessive amount of redundant traffic and exaggerates interference in the shared medium among the neighboring nodes. In order to mitigate these problems, this paper proposes an efficient flooding scheme with directional antennas, called BDF (Bi-Directional Flooding). The BDF scheme limits the propagation range of the packet forwarding to two directions over four, in order to minimize an overlapped area due to omnidirectional flooding. To analyze its performance, BDF has been compared with omnidirectional flooding and NEDF schemes.

A New Feature Selection Method for Improving the Precision of Diagnosing Abnormal Protein Sequences by Support Vector Machine and Vectorization Method

Pattern recognition and classification problems are most popular issue in machine learning, and it seem that they meet their second golden age with bioinformatics. However, the dataset of bioinformatics has several distinctive characteristics compared to the data set in classical pattern recognition and classification research area. One of the most difficulties using this theory in bioinformatics is that raw data of DNA or protein sequences cannot be directly used as input data for machine learning because every sequence has different length of its own code sequences. Therefore, this paper introduces one of the methods to overcome this difficulty, and also argues that the capability of generalization in this method is very poor as showing simple experiments. Finally, this paper suggests different approach to select the fixed number of effective features by using Support Vector Machine, and noise whitening method. This paper also defines the criteria of this suggested method and shows that this method improves the precision of diagnosing abnormal protein sequences with experiment of classifying ovarian cancer data set.

An Efficient Resource Reservation Scheme using a Dynamic Branch Router for Mobile IPv6 Networks

RSVP (Resource reSerVation Protocol) is the Internet standard protocol for supporting QoS (Quality of Service) requirements by reserving network resources between the sender and the receiver. Several problems, such as common path identification and resource pre-reservation, should be solved to apply RSVP in wireless mobile environments. To resolve these problems, this paper proposes DBRA (Dynamic Branch Router Approach) based on two designated entities, candidate access routers and a branch router. While several RSVP sessions between a branch router and candidate routers are managed, DBRA makes use of only one RSVP session between the sender and a branch router. Based on a network simulation, the proposed scheme has been compared with MRSVP (Mobile RSVP) and HMRSVP (Hierarchical MRSVP) in terms of the average packet transmission rate, bandwidth utilization and RSVP session failure rate.

A new approach for finding an optimal solution and regularization by learning dynamic momentum

Regularization and finding optimal solution for the classification problems are well known issue in the machine learning, but most of researches have been separately studied or considered as a same problem about these two issues. However, it is obvious that these approaches are not always possible because the evaluation of the performance in classification problems is mostly based on the data distribution and learning methods; therefore this paper suggests a new approach to simultaneously deal with finding optimal regularization parameter and solution in classification and regression problems by introducing dynamically rescheduled momentum with modified SVM in kernel space.

An effective real time update rule for improving performances both the classification and regression problems in kernel methods

It is general solution to get an answer from both classification and regression problem that information in real world matches matrices. This paper treats primary space as a real world, and dual space that primary spaces transfers to new matrices using kernel. In practical study, there are two kinds of problems, complete system which can get an answer using inverse matrix and ill-posed system or singular system which cannot get an answer directly from inverse of the given matrix. Furthermore, the problems are often given by the latter condition; therefore, it is necessary to find regularization parameter to change ill-posed or singular problems into complete system. This paper compares each performance under both classification and regression problems among GCV, L-curve, and kernel methods. This paper also suggests dynamic momentum which is learning under the limited proportional condition between learning epoch and the performance of given problems to increase performance and precision for regularization. Finally, this paper shows the results that suggested solution can get better or equivalent results compared with GCV, and L-curve through the experiments using Iris data which are used to consider standard data in classification, Gaussian data which are typical data for singular system, and Shaw data which is an one-dimension image restoration problem.

Improving Learning Performance of Support Vector Machine using the Kernel Relaxation and the Dynamic Momentum

This paper proposes learning performance improvement of support vector machine using the kernel relaxation and the dynamic momentum. The dynamic momentum is reflected to different momentum according to current state. While static momentum is equally influenced on the whole, the proposed dynamic momentum algorithm can control to the convergence rate and performance according to the change of the dynamic momentum by training. The proposed algorithm has been applied to the kernel relaxation as the new sequential learning method of support vector machine presented recently. The proposed algorithm has been applied to the SONAR data which is used to the standard classification problems for evaluating neural network. The simulation results of proposed algorithm have better the convergence rate and performance than those using kernel relaxation and static momentum, respectively.