• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제5권1호
• /
• pp.157-176
• /
• 2011

An interest-based P2P constructs the peer connections based on similarities for efficient search of resources. A clustering technique using peer similarities as data is an effective approach to group the most relevant peers. However, the separation of groups produced from clustering lowers the scalability of a P2P network. Moreover, the interest-based approach is only concerned with user-level grouping where topology-awareness on the physical network is not considered. This paper proposes an efficient scalable search for the interest-based P2P system. A scalable multi-ring (SMR) based on fuzzy clustering handles the grouping of relevant peers and the proposed scalable search utilizes the SMR for scalability of peer queries. In forming the multi-ring, a minimized route function is used to determine the shortest route to connect peers on the physical network. Performance evaluation showed that the SMR acquired an accurate peer grouping and improved the connectivity rate of the P2P network. Also, the proposed scalable search was efficient in finding more replicated files throughout the peer network compared to other traditional P2P approaches.

• 로봇학회논문지
• /
• 제14권1호
• /
• pp.40-49
• /
• 2019

Reinforcement learning has been applied to various problems in robotics. However, it was still hard to train complex robotic manipulation tasks since there is a few models which can be applicable to general tasks. Such general models require a lot of training episodes. In these reasons, deep neural networks which have shown to be good function approximators have not been actively used for robot manipulation task. Recently, some of these challenges are solved by a set of methods, such as Guided Policy Search, which guide or limit search directions while training of a deep neural network based policy model. These frameworks are already applied to a humanoid robot, PR2. However, in robotics, it is not trivial to adjust existing algorithms designed for one robot to another robot. In this paper, we present our implementation of Guided Policy Search to the robotic arms of the Baxter Research Robot. To meet the goals and needs of the project, we build on an existing implementation of Baxter Agent class for the Guided Policy Search algorithm code using the built-in Python interface. This work is expected to play an important role in popularizing robot manipulation reinforcement learning methods on cost-effective robot platforms.

• 방송공학회논문지
• /
• 제26권7호
• /
• pp.855-861
• /
• 2021

Neural Architecture Search (NAS) is cutting-edge technology in the machine learning community. NAS Without Training (NASWOT) recently has been proposed to tackle the high demand of computational resources in NAS by leveraging some indicators to predict the performance of architectures before training. The advantage of these indicators is that they do not require any training. Thus, NASWOT reduces the searching time and computational cost significantly. However, NASWOT only considers high-performing networks which does not guarantee a fast inference speed on hardware devices. In this paper, we propose a multi objectives reward function, which considers the network's latency and the predicted performance, and incorporate it into the Reinforcement Learning approach to search for the best networks with low latency. Unlike other methods, which use FLOPs to measure the latency that does not reflect the actual latency, we obtain the network's latency from the hardware NAS bench. We conduct extensive experiments on NAS-Bench-201 using CIFAR-10, CIFAR-100, and ImageNet-16-120 datasets, and show that the proposed method is capable of generating the best network under latency constrained without training subnetworks.

• Current Optics and Photonics
• /
• 제6권2호
• /
• pp.161-170
• /
• 2022

Three-dimensional (3D) geometric models are introduced to correct vignetting, and a downhill simplex search is applied to determine the coefficients of a 3D model used in digital microscopy. Vignetting is nonuniform illuminance with a geometric regularity on a two-dimensional (2D) image plane, which allows the illuminance distribution to be estimated using 3D models. The 3D models are defined using generalized polynomials and arbitrary coefficients. Because the 3D models are nonlinear, their coefficients are determined using a simplex search. The cost function of the simplex search is defined to minimize the error between the 3D model and the reference image of a standard white board. The conventional and proposed methods for correcting the vignetting are used in experiments on four inspection systems based on machine vision and microscopy. The methods are investigated using various performance indices, including the coefficient of determination, the mean absolute error, and the uniformity after correction. The proposed method is intuitive and shows performance similar to the conventional approach, using a smaller number of coefficients.

• Steel and Composite Structures
• /
• 제45권4호
• /
• pp.579-590
• /
• 2022

One of the biggest problems in structural steel calculation is the design of structures using the lowest possible material weight, making this a slow and costly process. To achieve this objective, several optimization methods have been developed and tested. Nevertheless, a method that performs very efficiently when applied to different problems is not yet available. Based on this assumption, this work proposes a hybrid metaheuristic algorithm for geometric and dimensional optimization of space trusses, called Simulated Squirrel Search Algorithm, which consists of an association of the well-established neighborhood shifting algorithm (Simulated Annealing) with a recently developed promising population algorithm (Squirrel Search Algorithm, or SSA). In this study, two models are tried, being respectively, a classical model from the literature (25-bar space truss) and a roof system composed of space trusses. The structures are subjected to resistance and displacement constraints. A penalty function using Fuzzy Logic (FL) is investigated. Comparative analyses are performed between the Squirrel Search Algorithm (SSSA) and other optimization methods present in the literature. The results obtained indicate that the proposed method can be competitive with other heuristics.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2003년도 ICCAS
• /
• pp.61-64
• /
• 2003

A band function model paired comparison method (BMPC method) is a kind of a paired comparison methods. Considering the human ambiguities, the BMPC method expressing the human judgment characteristics as a monotonous increase function with some width. Since function types are not specified in a BMPC method, the solution is obtained from inequalities, and the solution is given as a domain. To solve the simultaneous inequalities, the sequential renew method is used in the previous BMPC method. However, the sequential renew method requires much computational effort and memories. Generally, in BMPC method, it is able to solve only a paired comparison table which has less 12-13 samples. For that purpose, a new fast solution algorithm is required. In this paper, we proposed a new “search vector method” which renews the solution domain without creating new edge vectors. By using the method, it is able to decrease the necessary memory spaces and time to solve. The proposed method makes it able to solve more than 15 samples paired comparison inspections which are impossible to solve by previous method.

• 한국전문물리치료학회지
• /
• 제22권2호
• /
• pp.30-40
• /
• 2015

Robot-assisted rehabilitation therapy has been used to increase physical function in post-stroke patients. The aim of this meta-analysis was to identify whether robot-assisted gait training can improve patients' functional abilities. A comprehensive search was performed of PubMed, Cochrane Central Register of Controlled Trials (CENTRAL), Physiotherapy Evidence Database (PEDro), Academic Search Premier (ASP), ScienceDirect, Korean Studies Information Service System (KISS), Research Information Sharing Service (RISS), Korea National Library, and the Korean Medical Database up to April, 2014. Fifteen eligible studies researched the effects of robot-assisted gait training to a control group. All outcome measures were classified by International Classification of Functioning, Disability, and Health (ICF) domains (body function and structures, activity, and participation) and were pooled for calculating the effect size. The overall effect size of the robot-assisted gait training was .356 [95% confidence interval (CI): .186~.526]. When the effect was compared by the type of electromechanical robot, Gait Trainer (GT) (.471, 95% CI: .320~.621) showed more effective than Lokomat (.169, 95% CI: .063~.275). In addition, acute stroke patients showed more improvement than others. Although robot-assisted gait training may improve function, but there is no scientific evidence about the appropriate treatment time for one session or the appropriate duration of treatment. Additional researchers are needed to include more well-designed trials in order to resolve these uncertainties.

• 한국음향학회지
• /
• 제20권8호
• /
• pp.58-66
• /
• 2001

본논문에서는 LFM (Linear Frequency Modulated) 신호를 사용하는 능동소나에서 적은 연산량으로 표적반사신호의 시간지연과 도플러를 추정하는 기법을 제안하였다. 제안한 기법에서는 일반적인 추정기법들이 가지는 연산량의 문제를 해결하기 위해 LFM 신호의 상호모호함수 (cross ambiguity function)에서 시간지연과 도플러의 관계를 나타내는 대수적인 관계식을 이용하였다. FML (Fast Maximum Likelihood) 기법을 기반으로 하여 시간지연과 도플러의 대수적 관계식을 유도하였으며, 이를 이용하여 일반적인 2차원 탐색 대신 2번의 1차원 탐색으로 시간지연과 도플러를 추정하였다. 다양한 신호대 잡음비 (SNR)에서 제안한 알고리즘의 추정오차를 분석하였으며, 제안한 알고리즘이 우수한 추정 성능을 보임을 확인하였다.

• 방송공학회논문지
• /
• 제18권5호
• /
• pp.669-679
• /
• 2013

본 논문에서는 실시간으로 스테레오 정합을 수행하기 위한 VLSI(Very Large Scale Integrated Circuit)구조를 제안한다. 스테레오 정합의 연산을 분석하여 중간 연산 결과를 재사용하여 연산량과 메모리 접근수를 최소화한다. 이러한 동작을 수행할 수 있는 스테레오 정합 연산 셀의 구조를 제안하고, 이를 병렬적으로 확장하여 탐색 범위 내의 모든 비용함수를 동시에 연산할 수 있는 하드웨어의 구조를 제안한다. 이러한 하드웨어 구조를 확장하여 2차원 영역에 대한 비용함수를 연산할 수 있는 하드웨어의 구조와 동작을 제안한다. 구현한 하드웨어는 FPGA(Field Programmable Gate Array) 환경에서 최소 250Mhz의 클록 주파수에서 동작이 가능하고, 64화소의 탐색범위를 적용한 경우에 $640{\times}480$ 스테레오 영상을 약 805fps의 성능으로 처리할 수 있다.

• 한국통신학회논문지
• /
• 제24권7A호
• /
• pp.1080-1089
• /
• 1999

병렬 연쇄 컨볼루션 부호 및 직렬 연쇄 컨볼루션 부호의 ML 연판정 복호에 대한 비트오율확률의 상한치는 가중치 열거함수(Weight Enumerating Function; WEF)를 통해서 구할 수 있으며, 이 상한치는 반복 택 알고리듬과 양방향 탐색 알고리듬을 혼합한 새로운 오류사건 탐색 알고복호를 통해 얻을 수 있는 비트오류확률의 하한치가 된다. 본 논문에서는 스리듬을 제안하고, 얻어진 오류사건을 이용하여 WEF를 계산하는 알고리듬을 제안한다. 컴퓨터 시뮬레이션을 통해, 반복복호를 통해 얻을 수 있는 비트오류확률의 하한치가 됨을 확인하였다.