• International Journal of Computer Science & Network Security
• /
• v.21 no.6
• /
• pp.169-180
• /
• 2021

Since machine learning was invented, there have been many different machine learning-based algorithms, from shallow learning to deep learning models, that provide solutions to the classification tasks. But then it poses a problem in choosing a suitable classification algorithm that can improve the classification/detection efficiency for a certain network context. With that comes whether an algorithm provides good performance, why it works in some problems and not in others. In this paper, we present a data-centric analysis to provide a way for selecting a suitable classification algorithm. This data-centric approach is a new viewpoint in exploring relationships between classification performance and facts and figures of data sets.

• 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.

• Geomechanics and Engineering
• /
• v.32 no.3
• /
• pp.271-291
• /
• 2023

Settlement estimation in cohesion materials is a crucial topic to tackle because of the complexity of the cohesion soil texture, which could be solved roughly by substituted solutions. The goal of this research was to implement recently developed machine learning features as effective methods to predict settlement (S_m) of shallow foundations over cohesion soil properties. These models include hybridized support vector regression (SVR), random forests (RF), and coot optimization algorithm (COM), and black widow optimization algorithm (BWOA). The results indicate that all created systems accurately simulated the S_m, with an R² of better than 0.979 and 0.9765 for the train and test data phases, respectively. This indicates extraordinary efficiency and a good correlation between the experimental and simulated S_m. The model's results outperformed those of ANFIS - PSO, and COM - RF findings were much outstanding to those of the literature. By analyzing established designs utilizing different analysis aspects, such as various error criteria, Taylor diagrams, uncertainty analyses, and error distribution, it was feasible to arrive at the final result that the recommended COM - RF was the outperformed approach in the forecasting process of S_m of shallow foundation, while other techniques were also reliable.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.66 no.7
• /
• pp.1100-1104
• /
• 2017

A detector-based approach shows the limited performances for the defect inspections such as shallow fine cracks and indistinguishable defects from background. Deep learning technique is widely used for object recognition and it's applications to detect defects have been gradually attempted. Deep learning requires huge scale of learning data, but acquisition of data can be limited in some industrial application. The possibility of applying CNN which is one of the deep learning approaches for surface defect inspection is investigated for industrial parts whose detection difficulty is challenging and learning data is not sufficient. VOV is adopted for pre-processing and to obtain a resonable number of ROIs for a data augmentation. Then CNN method is applied for the classification. Three CNN networks, AlexNet, VGGNet, and mofified VGGNet are compared for experiments of defects detection.

• Journal of Information Processing Systems
• /
• v.17 no.2
• /
• pp.385-398
• /
• 2021

The complexity of deep learning models affects the real-time performance of gesture recognition, thereby limiting the application of gesture recognition algorithms in actual scenarios. Hence, a residual learning neural network based on a deep convolutional neural network is proposed. First, small convolution kernels are used to extract the local details of gesture images. Subsequently, a shallow residual structure is built to share weights, thereby avoiding gradient disappearance or gradient explosion as the network layer deepens; consequently, the difficulty of model optimisation is simplified. Additional convolutional neural networks are used to accelerate the refinement of deep abstract features based on the spatial importance of the gesture feature distribution. Finally, a fully connected cascade softmax classifier is used to complete the gesture recognition. Compared with the dense connection multiplexing feature information network, the proposed algorithm is optimised in feature multiplexing to avoid performance fluctuations caused by feature redundancy. Experimental results from the ISOGD gesture dataset and Gesture dataset prove that the proposed algorithm affords a fast convergence speed and high accuracy.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.157-157
• /
• 2018

A reliable and accurate downscaling model which can provide climate change information, obtained from global climate models (GCMs), at finer resolution has been always of great interest to researchers. In order to achieve this model, linear methods widely have been studied in the past decades. However, nonlinear methods also can be potentially beneficial to solve downscaling problem. Therefore, this study explored the applicability of some nonlinear machine learning techniques such as neural network (NN), extreme learning machine (ELM), and ELM autoencoder (ELM-AE) as well as a linear method, least absolute shrinkage and selection operator (LASSO), to build a reliable temperature downscaling model. ELM is an efficient learning algorithm for generalized single layer feed-forward neural networks (SLFNs). Its excellent training speed and good generalization capability make ELM an efficient solution for SLFNs compared to traditional time-consuming learning methods like back propagation (BP). However, due to its shallow architecture, ELM may not capture all of nonlinear relationships between input features. To address this issue, ELM-AE was tested in the current study for temperature downscaling.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.2
• /
• pp.327-329
• /
• 2021

Underwater acoustic communication channel is influenced by environmental parameters such as multipath, background noise and scattering. Therefore, a transmitted signal is influenced by the sea surface and the sea bottom boundaries, and a received signal shows a delay spread. These factors create a noise in the image and degrade the quality of underwater acoustic communication. To solve these problems, in this paper, we evaluate the performance of an underwater acoustic communication model using a denoising auto-encoder used for unsupervised learning. Noise images generated by the underwater multipath channel were collected and used as training data. Experimental results were analyzed as a PSNR parameter that expressed the noise ratio of the two images.

• Hwankyungkyoyuk
• /
• v.20 no.1
• /
• pp.19-27
• /
• 2007

This study was intended to answer the question, 'What kinds of effects will be aroused by experiential learning for environment based on living area?'. Experiential learning for environment was operated to 17 elementary school students in 4th grade in Kyeong-san city. The results were drawn analyzing the mind map for the changes of environmental consciousness before and after learning, and they are as below. First, it had an effect to change the meaning association of the relationship between 'river and me'. Meaning association was 'river-a thing' before experiential learning, but it was developed as 'river-a thing-me' after learning. This means that students expanded understanding of the world that they were belonging and self-spatialization was promoted. The expansion of meaning association would be a start point and a method to promote their segmentation for each student. Second, students could self-directly modify misconception and preconception after experiential learning. It showed that students could find meanings in the world that they were belonging by experiential learning for environment, and misconception obtained by concept learning without actual situation could be revised through the truth recognition in meanings, and student could see what things displayed. Therefore preconception would be corrected. Of course, everything would not be completed by just one time of experiential learning, and consistent experience learning should be operated. Third, experiential learning promoted the change of sensitivity. Students had shallow sensitivity, which appeared in the relation with things, since having learned only inside of class without a direct observation. However their sensitivity could be increased by experiencing specific things. Fourth, there was the change of classification recognition. Students found properties of things with a direct observation. It raised their ability to classify things, and to understand an individual thing in 'a class'.

• Journal of Digital Convergence
• /
• v.19 no.6
• /
• pp.251-258
• /
• 2021

In this paper, we proposed an encoder-decoder model utilizing residual learning to improve the accuracy of the U-Net-based semantic segmentation method. U-Net is a deep learning-based semantic segmentation method and is mainly used in applications such as autonomous vehicles and medical image analysis. The conventional U-Net occurs loss in feature compression process due to the shallow structure of the encoder. The loss of features causes a lack of context information necessary for classifying objects and has a problem of reducing segmentation accuracy. To improve this, The proposed method efficiently extracted context information through an encoder using residual learning, which is effective in preventing feature loss and gradient vanishing problems in the conventional U-Net. Furthermore, we reduced down-sampling operations in the encoder to reduce the loss of spatial information included in the feature maps. The proposed method showed an improved segmentation result of about 12% compared to the conventional U-Net in the Cityscapes dataset experiment.

• International Journal of Advanced Culture Technology
• /
• v.6 no.4
• /
• pp.12-19
• /
• 2018

This article aims to discuss the transition of the Kazakh writing system from Cyrillic to Latin. First, the study investigates the relationship between the Kazakh Cyrillic alphabet and phonology, in order to linguistically evaluate the efficiency of the writing system. Second, the process of determining the Kazakh Latin alphabet is discussed in terms of the Kazakh phonological system. Third, the factors that determined the Latin alphabet of Kazakh language are analyzed. In Kazakh, the phonemic system is subject to controversy among linguists, but it can be said that the phonological system basically follows the one-to-one correspondence to the Russian and Kazakh phonemes. As for the depth of orthographies, Kazakh Cyrillic writing system is not based on the shallow orthographies, so it incorporates morphophonemic information to make skilled readers understand easier. The political and social aspects are considered as a cause of the alphabet change. Although there are studies suggesting the conversion of the writing system is caused by the extrinsic factors rather than the intrinsic factors, the five criteria of Smalley (1964), which compromise the intrinsic and extrinsic factors, are also persuasive. The five factors are 1) Maximum motivation for the learner, 2) Maximum representation of speech, 3) Maximum ease of learning, 4) Maximum transfer, 5) Maximum ease of reproduction.