• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.311-315
• /
• 2021

근래의 자연어처리 분야에서는 잘 만들어진 도구(Library)를 이용하여 생산성 높은 개발과 연구가 활발하게 이뤄지고 있다. 이 중에 대다수는 깊은 학습(Deep-Learning, 딥러닝) 기반인데, 이런 모델들은 학습 속도가 느리고, 비용이 비싸고, 사용(Run-Time) 속도도 느리다. 이뿐만 아니라 라벨(Label)의 가짓수가 굉장히 많거나, 라벨의 구성이 단어마다 달라질 수 있는 의미분별(동형이의어, 다의어 번호 태깅) 분야에서 딥러닝은 굉장히 비효율적인 문제가 있다. 이런 문제들은 오히려 기존의 얕은 학습(Shallow-Learning)기반 모델에서는 없던 것들이지만, 최근의 연구경향에서 딥러닝 비중이 급격히 증가하면서, 멀티스레딩 같은 고급 기능들을 지원하는 얕은 학습 기반 언어모델이 새로이 개발되지 않고 있었다. 본 논문에서는 학습과 태깅 모두에서 멀티스레딩을 지원하고, 딥러닝에서 연구된 드롭아웃 기법이 구현된 자연어처리 도구인 혼합 자질 가변 표지기 ManiFL(Manifold Feature Labelling : ManiFL)을 소개한다. 본 논문은 실험을 통해서 ManiFL로 다의어태깅이 가능함을 보여주고, 딥러닝과 CRFsuite에서 높은 성능을 보여주는 개체명 인식에서도 비교할만한 성능이 나옴을 보였다.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.16 no.1
• /
• pp.291-299
• /
• 2016

This paper raises a question of whether a simple periodic phenomenon is associated with the topology and provides the convincing answers to it. A variety of music instrumental sound signals are used to prove our assertion, which are embedded in Euclidean space to analyze their topologies by computing the homology groups. A commute time embedding is employed to transform segments of waveforms into the corresponding geometries, which is implemented by organizing patches according to the graph-based metric. It is shown that commute time embedding generates the intrinsic topological complexities although their geometries are varied according to the spectrums of the signals. This paper employs a persistent homology to determine the topological invariants of the simplicial complexes constructed by randomly sampling the commute time embedding of the waveforms, and discusses their applications.

• Journal of Korea Multimedia Society
• /
• v.19 no.2
• /
• pp.146-154
• /
• 2016

This paper presents methods to analyze the topological structures of the spaces composed of patches extracted from waveform signals, which can be applied to the classification of signals. Commute time embedding is performed to transform the patch sets into the corresponding geometries, which has the properties that the embedding geometries of periodic or quasi-periodic waveforms are represented as closed curves on the low dimensional Euclidean space, while those of aperiodic signals have the shape of open curves. Persistent homology is employed to determine the topological invariants of the simplicial complexes constructed by randomly sampling the commute time embedding of the waveforms, which can be used to discriminate between the groups of waveforms topologically.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.07b
• /
• pp.748-750
• /
• 2005

Isomap [1] is a manifold learning algorithm, which extends classical multidimensional scaling (MDS) by considering approximate geodesic distance instead of Euclidean distance. The approximate geodesic distance matrix can be interpreted as a kernel matrix, which implies that Isomap can be solved by a kernel eigenvalue problem. However, the geodesic distance kernel matrix is not guaranteed to be positive semidefinite. In this paper we employ a constant-adding method, which leads to the Mercer kernel-based Isomap algorithm. Numerical experimental results with noisy 'Swiss roll' data, confirm the validity and high performance of our kernel Isomap algorithm.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2018.10a
• /
• pp.763-766
• /
• 2018

표정은 사람들 사이에서 감정을 표현하는 강력한 비언어적 수단이다. 표정 인식은 기계학습에서 아주 중요한 분야 중에 하나이다. 표정 인식에 사용되는 기계학습 모델들은 사람 수준의 성능을 보여준다. 하지만 좋은 성능에도 불구하고, 기계학습 모델들은 표정 인식 결과에 대한 근거나 설명을 제공해주지 못한다. 이 연구는 표정 인식의 근거로서 Facial Action Coding Unit(AUs)을 사용하기 위해서 CK+ Dataset을 사용하여 표정 인식을 학습한 Convolutional Neural Network(CNN) 모델이 추출한 특징들을 t-distributed stochastic neighbor embedding(t-SNE)을 사용하여 시각화한 뒤, 인식된 표정과 AUs 사이의 분포의 연관성을 확인하는 연구이다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.10a
• /
• pp.39-42
• /
• 2006

뉴럴네트워크에서 학습은 에러를 줄이는 방법으로 구현 된다. 이 때 parameter 공간에서 Risk function은 multi-minima potential로 표현 될 수 있으며 우리의 목적은 global minimum weight 좌표를 얻는 것이다. 이전의 연구로는 Attouch et al.의 damped oscillator 방정식을 이용한 방법이 있고, Qian의 critically damped oscillator를 통한 steepest descent의 momentum과 learning parameter 유도가 있다. 우리는 이 두 연구를 참고로 manifold 상에서 최단 경로인 geodesic을 Newton 역학의 Lagrangian에 적용함으로써 adaptive steepest descent 학습법을 얻었다. 우리는 이 새로운 방법을 Rosenbrock 과 Griewank 포텐셜들에 적용하여 그 성능을 알아 본다.

• The Journal of the Korea Contents Association
• /
• v.9 no.8
• /
• pp.32-40
• /
• 2009

This research provides a new approach for implementing an educational content for Historic Education in order to provide an effective learning environment. From historic educational point of view, it is important to comprehend a historical fact in the context of the situation at that time. So, this paper suggests that the historic content should describe not only information about various relics and ruins but also historical relationship and background. In this system, we provide versatile type of contents to help learners for collecting manifold informations about their interesting era. And this system proffers natural and residential 3D environments, which give learners to understand conceivably and to think collectively. Using the interactions, the learners navigating this virtual world are able to construct their own information system through selecting a interested one among the offered contents in the system and consequently they are getting a scientific thinking power and a creative imagination.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.12
• /
• pp.2946-2952
• /
• 2014

Among the Example based Super Resolution(SR) techniques, Neighbor embedding(NE) has been inspired by manifold learning method, particularly locally linear embedding. However, the poor generalization of NE decreases the performance of such algorithm. The sizes of local training sets are always too small to improve the performance of NE. We propose the Learning Sparse-Neighbor Image Representation baesd on SVR having an excellent generalization ability to solve this problem. Given a low resolution image, we first use bicubic interpolation to synthesize its high resolution version. We extract the patches from this synthesized image and determine whether each patch corresponds to regions with high or low spatial frequencies. After the weight of each patch is obtained by our method, we used to learn separate SVR models. Finally, we update the pixel values using the previously learned SVRs. Through experimental results, we quantitatively and qualitatively confirm the improved results of the proposed algorithm when comparing with conventional interpolation methods and NE.

• Journal of the Korean Geotechnical Society
• /
• v.39 no.4
• /
• pp.5-17
• /
• 2023

The advancement in large-scale underground excavation in urban areas necessitates monitoring and predicting technologies that can pre-emptively mitigate risk factors at construction sites. Traditionally, two methods predict the deformation of retaining walls induced by excavation: empirical and numerical analysis. Recent progress in artificial intelligence technology has led to the development of a predictive model using machine learning techniques. This study developed a model for predicting the deformation of a retaining wall under construction using a boosting-based algorithm and an ensemble model with outstanding predictive power and efficiency. A database was established using the data from the design-construction-maintenance process of the underground retaining wall project in a manifold manner. Based on these data, a learning model was created, and the performance was evaluated. The boosting and ensemble models demonstrated that wall deformation could be accurately predicted. In addition, it was confirmed that prediction results with the characteristics of the actual construction process can be presented using data collected from ground measurements. The predictive model developed in this study is expected to be used to evaluate and monitor the stability of retaining walls under construction.

• Journal of IKEEE
• /
• v.23 no.4
• /
• pp.1295-1301
• /
• 2019

In this paper, we propose a novel method of denoising human motion using a bidirectional recurrent neural network (BRNN) with an attention mechanism. The corrupted motion captured from a single 3D depth sensor camera is automatically fixed in the well-established smooth motion manifold. Incorporating an attention mechanism into BRNN achieves better optimization results and higher accuracy than other deep learning frameworks because a higher weight value is selectively given to a more important input pose at a specific frame for encoding the input motion. Experimental results show that our approach effectively handles various types of motion and noise, and we believe that our method can sufficiently be used in motion capture applications as a post-processing step after capturing human motion.