• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권5호
• /
• pp.2381-2399
• /
• 2019

Traditional recommendation algorithms on Collaborative Filtering (CF) mainly focus on the rating prediction with explicit ratings, and cannot be applied to the top-N recommendation with implicit feedbacks. To tackle this problem, we propose a new collaborative filtering approach namely Maximize MAP with Matrix Factorization (MFMAP). In addition, in order to solve the problem of non-smoothing loss function in learning to rank (LTR) algorithm based on pairwise, we also propose a smooth MAP measure which can be easily implemented by standard optimization approaches. We perform experiments on three different datasets, and the experimental results show that the performance of MFMAP is significantly better than other recommendation approaches.

• 한국정밀공학회지
• /
• 제12권7호
• /
• pp.32-45
• /
• 1995

In this paper, we present a new method to tool path planning problem for rough cut of pocket milling operations. The key idea is to formulate the tool path problem into a TSP (Travelling Salesman Problem) so that the powerful neural network approach can be effectively applied. Specifically, our method is composed of three procedures: a) discretization of the pocket area into a finite number of tool points, b) neural network approach (called SOM-Self Organizing Map) for path finding, and c) postprocessing for path smoothing and feedrate adjustment. By the neural network procedure, an efficient tool path (in the sense of path length and tool retraction) can be robustly obtained for any arbitrary shaped pockets with many islands. In the postprocessing, a) the detailed shape of the path is fine tuned by eliminating sharp corners of the path segments, and b) any cross-overs between the path segments and islands. With the determined tool path, the feedrate adjustment is finally performed for legitimate motion without requiring excessive cutting forces. The validity and powerfulness of the algorithm is demonstrated through various computer simulations and real machining.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2021년도 학술발표회
• /
• pp.145-145
• /
• 2021

Water distribution networks (WDNs) are designed to satisfy water requirement of an urban community. One of the central issues in human history is providing sufficient quality and quantity of water through WDNs. A WDN consists of a great number of pipelines with different ages, lengths, materials, and sizes in varying degrees of deterioration. The available annual budget for rehabilitation of these infrastructures only covers part of the network; thus it is important to manage the limited budget in the most cost-effective manner. In this study, a novel pipe replacement scheduling approach is proposed in order to smooth the annual investment time series based on a life cycle cost assessment. The proposed approach is applied to a real WDN currently operating in South Korea. The proposed scheduling plan considers both the annual budget limitation and the optimum investment on pipes' useful life. A non-dominated sorting genetic algorithm is used to solve a multi-objective optimization problem. Three decision-making objectives, including the minimum imposed LCC of the network, the minimum standard deviation of annual cost, and the minimum average age of the network, are considered to find optimal pipe replacement planning over long-term time period. The results indicate that the proposed scheduling structure provides efficient and cost-effective rehabilitation management of water network with consistent annual budget.

• 한국압력기기공학회 논문집
• /
• 제19권2호
• /
• pp.146-154
• /
• 2023

In nuclear power plants, ultrasonic test for pipe wall thickness measurement is used during periodic inspections to prevent pipe rupture due to pipe wall thinning. However, when measuring pipe wall thickness using ultrasonic test, a significant amount of measurement error occurs due to the on-site conditions of the nuclear power plant. If the maximum pipe wall thinning rate is decided by the measured pipe wall thickness containing a significant error, the pipe wall thinning rate data have significant uncertainty and systematic overestimation. This study proposes preprocessing of pipe wall thinning measurement data using support vector machine regression algorithm. By using support vector machine, pipe wall thinning measurement data can be smoothened and accordingly uncertainty and systematic overestimation of the estimated pipe wall thinning rate data can be reduced.

• 로봇학회논문지
• /
• 제19권3호
• /
• pp.236-243
• /
• 2024

This paper introduces a dual manipulator system designed to automate the packaging process of industrial coils, which exhibit higher variability than other structured industrial fields due to diverse commercial requirements. The conventional solution involves the direct-teaching method, where an operator instructs the robot on a target configuration. However, this method has distinct limitations, such as low flexibility in dealing with varied sizes and safety concerns for the operators handling large products. In this sense, this paper proposes a two-step approach for coil packaging: motion planning and assembly execution. The motion planning includes a Rapidly-exploring Random Tree algorithm and a smoothing method, allowing the robot to reach the target configuration. In the assembly execution, the packaging is considered a peg-in-hole assembly. Unlike typical peg-in-hole assembly handling two workpieces, the packaging includes three workpieces (e.g., coil, inner ring, side plate). To address this assembly, the paper suggests a suitable strategy for dual manipulation. Finally, the validity of the proposed system is demonstrated through experiments with three different sizes of coils, replicating real-world packaging situations.

• Journal of Information Processing Systems
• /
• 제20권4호
• /
• pp.501-513
• /
• 2024

In recent years, graph neural networks (GNNs) have been extensively used to analyze graph data across various domains because of their powerful capabilities in learning complex graph-structured data. However, recent research has focused on improving the performance of a single GNN with only two or three layers. This is because stacking layers deeply causes the over-smoothing problem of GNNs, which degrades the performance of GNNs significantly. On the other hand, ensemble methods combine individual weak models to obtain better generalization performance. Among them, gradient boosting is a powerful supervised learning algorithm that adds new weak models in the direction of reducing the errors of the previously created weak models. After repeating this process, gradient boosting combines the weak models to produce a strong model with better performance. Until now, most studies on GNNs have focused on improving the performance of a single GNN. In contrast, improving the performance of GNNs using multiple GNNs has not been studied much yet. In this paper, we propose gradient boosted graph neural networks (GBGNN) that combine multiple shallow GNNs with gradient boosting. We use shallow GNNs as weak models and create new weak models using the proposed gradient boosting-based loss function. Our empirical evaluations on three real-world datasets demonstrate that GBGNN performs much better than a single GNN. Specifically, in our experiments using graph convolutional network (GCN) and graph attention network (GAT) as weak models on the Cora dataset, GBGNN achieves performance improvements of 12.3%p and 6.1%p in node classification accuracy compared to a single GCN and a single GAT, respectively.

• 한국정보통신학회논문지
• /
• 제22권2호
• /
• pp.276-283
• /
• 2018

기후변화 등으로 해안 침수 등의 피해가 증가하고 있으며, 이러한 피해를 줄이기 위해 해양을 지속적으로 모니터링하기 위한 연구들이 진행되고 있다. 본 논문에서는 해수면의 변화를 모니터링하고 위험한 상황을 경보하는 해양모니터링 시스템에 적용할 수 있는 조위 예측 모델을 제안한다. 기존의 조위 예측 모델은 장기적인 예보를 위한 것으로 많은 데이터와 복잡한 알고리즘이 필요하기 때문에 실시간 시스템에는 적절하지 않다. 반면, 제안하는 알고리즘은 조위 센서에 의해 측정된 데이터를 이용하여 실시간으로 조위를 예측하는 방법으로 간단하지만 정확하게 한 시간 또는 두 시간의 비교적 짧은 시간 후의 조위를 예측한다. 제안하는 방법은 조석의 조화분석을 위해 칼만필터 알고리즘을 사용하고 추가적인 오류 보정을 위해 이중 지수 평활법을 사용한다. 실험 결과는 제안하는 알고리즘이 간단하지만 정확하게 조위를 예측하는 것을 보여준다.

• 전자공학회논문지
• /
• 제50권6호
• /
• pp.276-286
• /
• 2013

Singular Value Decomposition (SVD) 기반의 디콘볼루션 방식은 CT 관류 영상 해석에서 가장 널리 사용되는 기법이다. 이 방식에서는 잡음의 영향을 줄이기 위해 SVD 계수를 임계화하는 과정이 사용된다. 이 때 임계화 경계치로 고정된 값을 사용하거나 미리 정해진 진동 지수(Oscillation Index)에 따른 경계치가 사용된다. 이들 두 임계화 방식은 계산량과 정확도 측면에서 서로 장단점을 가지고 있다. 본 논문에서는 두 임계화 방식의 정확도를 비교하기 위한 몬테 칼로 모의 실험 방식을 제안한다. 또한 관류 해석시 사용하는 평활화 과정이 알고리즘의 정확도에 미치는 영향을 측정하기 위해 이 실험 방식을 확장하였다. 본 논문에서는 이와 같은 성능 비교를 위한 모의 실험 방식을 제시하고, 모의 데이터와 실제 CT 영상에 대한 실험 결과를 소개한다.

• 방송공학회논문지
• /
• 제18권3호
• /
• pp.342-348
• /
• 2013

본 논문에서는 적응적인 확장 블록을 사용하는 프레임율 증강 기법인 AEBME (Adaptive Extended Bilateral Motion Estimation)을 제안하고자 한다. 기존의 EBME (Extended Bilateral Motion Estimation) 알고리듬은 동일한 구역에 두 번의 움직임 예측을 수행함으로 인해 높은 계산량이 요구되었다. 본 논문에서는 영상의 edge 정보를 활용한 블록 유형의 일치 유무를 고려하여 EBME 수행여부를 결정함으로써 움직임 예측 과정을 보다 빠르게 수행하도록 하였다. 움직임 벡터 평활화 과정이 적용되어 움직임 벡터 필드 내의 이상 벡터를 찾아 수정한다. 최종적으로 OBMC (Overlapped Block Motion Compensation)와 MCFI (Motion Compensated Frame Interpolation)이 프레임 움직임의 성질에 따라 적용되어 중간 프레임을 보간하게 된다. 실험 결과를 통해 제안하는 알고리듬이 기존의 알고리듬인 EBME에 비해 향상된 성능과 빠른 속도를 보임을 알 수 있다.

• 한국근적외분광분석학회:학술대회논문집
• /
• 한국근적외분광분석학회 2001년도 NIR-2001
• /
• pp.4102-4102
• /
• 2001

In multivariate analysis, absorbance spectrum is measured over a band of wavelengths. One does not often pay attention to the size of this wavelength band. However, it is desirable that spectrum is measured at only necessary wavelengths as long as the acceptable accuracy of prediction can be met. In this paper, the method of selecting an optimal band of wavelengths based on the loading vector analysis was proposed and applied for determining total protein in human serum using near-infrared transmission spectroscopy and PLSR. Loading vectors in the full spectrum PLSR were used as reference in selecting wavelengths, but only the first loading vector was used since it explains the spectrum best. Absorbance spectra of sera from 97 outpatients were measured at 1530∼1850 nm with an interval of 2 nm. Total protein concentrations of sera were ranged from 5.1 to 7.7 g/㎗. Spectra were measured by Cary 5E spectrophotometer (Varian, Australia). Serum in the 5 mm-pathlength cuvette was put in the sample beam and air in the reference beam. Full spectrum PLSR was applied to determine total protein from sera. Next, the wavelength region of 1672∼1754 nm was selected based on the first loading vector analysis. Standard Error of Cross Validation (SECV) of full spectrum (1530∼l850 nm) PLSR and selected wavelength PLSR (1672∼1754 nm) was respectively 0.28 and 0.27 g/㎗. The prediction accuracy between the two bands was equal. Wavelength selection based on loading vector in PLSR seemed to be simple and robust in comparison to other methods based on correlation plot, regression vector and genetic algorithm. As a reference of wavelength selection for PLSR, the loading vector has the advantage over the correlation plot since the former is based on multivariate model whereas the latter, on univariate model. Wavelength selection by the first loading vector analysis requires shorter computation time than that by genetic algorithm and needs not smoothing.