• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.141-145
• /
• 2022

다중 문서 기반 대화 시스템에서 응답 시스템은 올바른 답변을 생성하기 위해서 여러 개의 문서 중 질문과 가장 관련 있는 문서를 검색하는 것부터 시작해야 한다. DialDoc 2022 Shared Task[1]를 비롯한 최근의 연구들은 대화 시스템의 문서 검색 과정을 위해 Dense Passage Retrieval(DPR)[2] 모델을 사용하고 있으며 검색기의 성능 개선을 위해 Re-ranking과 Hard negative sampling 같은 방법들이 연구되고 있다. 본 논문에서는 문서에 기반하는 대화 데이터의 양이 적거나 제한될 경우, 주어진 데이터를 효율적으로 활용해 보고자 검색기를 생성 모델을 이용하여 문서의 엔티티를 기반으로 질문을 생성하고 기존 데이터에 증강하는 방법을 제시했으며 실험의 결과로 MRR metric의 경우 0.96 ~ 1.56의 성능 향상을, R@1 metric의 경우 1.2 ~ 1.57의 성능 향상을 확인하였다.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.48 no.6
• /
• pp.114-122
• /
• 2011

We propose an automatic localization technique based on Uniform Arc Length Sampling (UALS) of 360 degree range sensor data. The proposed method samples 3D points from dense a point-cloud which is acquired by the sensor, registers the sampled points to a digital surface model(DSM) in real-time, and determines the location of an Unmanned Ground Vehicle(UGV). To reduce the sampling and registration time of a sequence of dense range data, 3D range points are sampled uniformly in terms of ground sample distance. Using the proposed method, we can reduce the number of 3D points while maintaining their uniformity over range data. We compare the registration speed and accuracy of the proposed method with a conventional sample method. Through several experiments by changing the number of sampling points, we analyze the speed and accuracy of the proposed method.

• Journal of Sensor Science and Technology
• /
• v.29 no.3
• /
• pp.141-148
• /
• 2020

To find food, homes, and mates, some animals have adapted special sensing capabilities. Rather than using a passive method, they discharge a signal and then extract the necessary information from the response. More importantly, they use the slope of the detected signal to find the destination of an object. In this paper, similar strategy is mathematically formulated. A perturbation and correlation-based gradient estimation method is developed and used as a sensing strategy. This method allows us to adaptively sense an object in a given environment effectively. The proposed strategy is based on the use of gradient values; rather than instantaneous measurements. Considering the gradient value, the sampling frequency is planned adaptively, i.e., sparse sampling is performed in slowly varying regions, while dense sampling is conducted in rapidly changing regions. Using a temperature sensor, the proposed strategy is verified and its effectiveness is demonstrated.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.242-251
• /
• 2020

Recently, many deep convolutional neural networks for image super-resolution have been studied. Existing deep learning-based super-resolution algorithms are architecture that up-samples the resolution at the end of the network. The post-upsampling architecture has an inefficient structure at large scaling factor result of predicting a lot of information for mapping from low-resolution to high-resolution at once. In this paper, we propose a single image super-resolution using Channel Attention Residual Dense Block based on an iterative up-down sampling architecture. The proposed algorithm efficiently predicts the mapping relationship between low-resolution and high-resolution, and shows up to 0.14dB performance improvement and enhanced subjective image quality compared to the existing algorithm at large scaling factor result.

• 한국데이터정보과학회:학술대회논문집
• /
• 2003.10a
• /
• pp.249-258
• /
• 2003

K-means clustering has been widely used in many applications, such that pattern analysis or recognition, data analysis, image processing, market research and so on. It can identify dense and sparse regions among data attributes or object attributes. But k-means algorithm requires many hours to get k clusters that we want, because it is more primitive, explorative. In this paper we propose a new method of k-means clustering using the grid-based sample. It is more fast than any traditional clustering method and maintains its accuracy.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.361-365
• /
• 2023

오픈 도메인 질의응답 (ODQA, Open-Domain Question Answering)은 주어진 질문에 대한 답을 찾는 작업으로 일반적으로 질문과 관련 있는 지식을 검색 모델(Retrieval)을 통해 찾는 단계와, 찾은 지식에서 문서의 정답을 독해 모델(Reader)을 이용하여 찾는 단계로 구성되어 있다. 본 논문은 기존의 DPR(Dense Passage Retrieval)을 이용한 복수의 검색 모델(Retrieval)만을 계층적으로 사용하여 독해 모델(Reader)을 사용하지 않고 정답 문장을 찾는 방법과 정답 문장을 찾는 데 특화된 검색 모델 학습을 위한 유효한 성능 향상을 보이는 Hard Negative Sampling 기법을 제안한다. 해당 제안기법을 적용한 결과, 동일 조건에서 학습된 검색 - 독해(Retrieval-Reader) 구조의 베이스라인 모델보다 EM에서 12%, F1에서 10%의 성능 향상을 보였다.

• Journal of Electrical Engineering and Technology
• /
• v.7 no.5
• /
• pp.789-796
• /
• 2012

Fingerprinting is a widely used positioning technology for received signal strength (RSS) based wireless local area network (WLAN) positioning system. Though spatial RSS variation is the key factor of the positioning technology, temporal RSS variation needs to be considered for more accuracy. To deal with the spatial and temporal RSS characteristics within a unified framework, this paper proposes an extended signal propagation mode (ESPM) and a fingerprint generation method. The proposed spatiotemporal fingerprint generation method consists of two algorithms running in parallel; Kalman filtering at several measurement-sampling locations and Kriging to generate location fingerprints at dense reference locations. The two different algorithms are connected by the extended signal propagation model which describes the spatial and temporal measurement characteristics in one frame. An experiment demonstrates that the proposed method provides an improved positioning accuracy.

• Ocean and Polar Research
• /
• v.23 no.4
• /
• pp.419-424
• /
• 2001

The Antarctic clam Laternula elliptica, is one of the most representative benthic invertebrates in the Antarctic nearshore waters. Endemic to the Antarctic, L. elliptica is widely distributed around the Antarctica occurring as dense patches in shallow sheltered areas and exhibits high biomass. Despite its apparent ecological importance, L. elliptica has rarely been studied until recently probably due to difficulties in sampling in the ice-impacted waters. Recent studies have revealed various aspects of its ecology and physiology. In this review, some physiological and ecological characteristics of this species are discussed in relation to some prevailing features of its habitat environment, in particular physical instability of habitat substrates and extreme seasonality of food availability.

• Journal of Computational Design and Engineering
• /
• v.1 no.4
• /
• pp.223-232
• /
• 2014

This paper describes a variant of the extended Gaussian image based registration algorithm for point clouds with surface color information. The method correlates the distributions of surface normals for rotational alignment and grid occupancy for translational alignment with hue filters applied during the construction of surface normal histograms and occupancy grids. In this method, the size of the point cloud is reduced with a hue-based down sampling that is independent of the point sample density or local geometry. Experimental results show that use of the hue filters increases the registration speed and improves the registration accuracy. Coarse rigid transformations determined in this step enable fine alignment with dense, unfiltered point clouds or using Iterative Common Point (ICP) alignment techniques.

• Journal of the Korean Data and Information Science Society
• /
• v.14 no.3
• /
• pp.535-543
• /
• 2003

Cluster analysis has been widely used in many applications, such as pattern analysis or recognition, data analysis, image processing, market research on on-line or off-line and so on. Clustering can identify dense and sparse regions among data attributes or object attributes. But it requires many hours to get clusters that we want, because clustering is more primitive, explorative and we make many data an object of cluster analysis. In this paper we propose a new method of clustering using sample based on grid. It is more fast than any traditional clustering method and maintains its accuracy.