• KIISE Transactions on Computing Practices
• /
• v.21 no.4
• /
• pp.338-342
• /
• 2015

Recurrent Neural Network (RNN), a machine learning model which can handle time-series data, can possess more varied structures than a feed-forward neural network, since a RNN allows hidden-to-hidden connections. This research focuses on the network structure among hidden neurons, and discusses the information processing capability of RNN. Time-series learning potential and dynamics of RNNs are investigated upon several well-established network structure models. Hidden neuron network structure is found to have significant impact on the performance of a model, and the performance variations are generally correlated with the criticality of the network dynamics. Especially Preferential Attachment Network model showed an interesting behavior. These findings provide clues for performance improvement of the RNN.

• Journal of Electrical Engineering and Technology
• /
• v.12 no.1
• /
• pp.436-442
• /
• 2017

A novel method of Skin lesion segmentation based on the combination of Texture and Neural Network is proposed in this paper. This paper combines the textures of different pixels in the skin images in order to increase the performance of lesion segmentation. For segmenting skin lesions, a two-step process is done. First, automatic border detection is performed to separate the lesion from the background skin. This begins by identifying the features that represent the lesion border clearly by the process of Texture analysis. In the second step, the obtained features are given as input towards the Recurrent Echo state neural networks in order to obtain the segmented skin lesion region. The proposed algorithm is trained and tested for 862 skin lesion images in order to evaluate the accuracy of segmentation. Overall accuracy of the proposed method is compared with existing algorithms. An average accuracy of 98.8% for segmenting skin lesion images has been obtained.

• Investigative Magnetic Resonance Imaging
• /
• v.21 no.4
• /
• pp.223-232
• /
• 2017

Purpose: To report the use of multiband accelerated echo-planar imaging (EPI) for resting-state functional MRI (rs-fMRI) to achieve rapid high temporal resolution at 3T compared to conventional EPI. Materials and Methods: rs-fMRI data were acquired from 20 healthy right-handed volunteers by using three methods: conventional single-band gradient-echo EPI acquisition (Data 1), multiband gradient-echo EPI acquisition with 240 volumes (Data 2) and 480 volumes (Data 3). Temporal signal-to-noise ratio (tSNR) maps were obtained by dividing the mean of the time course of each voxel by its temporal standard deviation. The resting-state sensorimotor network (SMN) and default mode network (DMN) were estimated using independent component analysis (ICA) and a seed-based method. One-way analysis of variance (ANOVA) was performed between the tSNR map, SMN, and DMN from the three data sets for between-group analysis. P < 0.05 with a family-wise error (FWE) correction for multiple comparisons was considered statistically significant. Results: One-way ANOVA and post-hoc two-sample t-tests showed that the tSNR was higher in Data 1 than Data 2 and 3 in white matter structures such as the striatum and medial and superior longitudinal fasciculus. One-way ANOVA revealed no differences in SMN or DMN across the three data sets. Conclusion: Within the adapted metrics estimated under specific imaging conditions employed in this study, multiband accelerated EPI, which substantially reduced scan times, provides the same quality image of functional connectivity as rs-fMRI by using conventional EPI at 3T. Under employed imaging conditions, this technique shows strong potential for clinical acceptance and translation of rs-fMRI protocols with potential advantages in spatial and/or temporal resolution. However, further study is warranted to evaluate whether the current findings can be generalized in diverse settings.

• Journal of Electrical Engineering and Technology
• /
• v.11 no.2
• /
• pp.265-273
• /
• 2016

A water wall system is one of the most important components of a boiler in a thermal power plant, and it is a nonlinear Multi-Input and Multi-Output (MIMO) system, with 6 inputs and 3 outputs. Three models are developed and comp for the controller design, including a linear model, a multilayer feed-forward neural network (MFNN) model and an Echo State Network (ESN) model. First, the linear model is developed by linearizing a given nonlinear model and is analyzed as a function of the operating point. Second, the MFNN and the ESN are developed by using training data from the nonlinear model. The three models are validated using Matlab with nonlinear input-output data that was not used during training.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.8
• /
• pp.498-504
• /
• 2005

We introduce a Korean speech recognition platform (ECHOS) developed for education and research Purposes. ECHOS lowers the entry barrier to speech recognition research and can be used as a reference engine by providing elementary speech recognition modules. It has an easy simple object-oriented architecture, implemented in the C++ language with the standard template library. The input of the ECHOS is digital speech data sampled at 8 or 16 kHz. Its output is the 1-best recognition result. N-best recognition results, and a word graph. The recognition engine is composed of MFCC/PLP feature extraction, HMM-based acoustic modeling, n-gram language modeling, finite state network (FSN)- and lexical tree-based search algorithms. It can handle various tasks from isolated word recognition to large vocabulary continuous speech recognition. We compare the performance of ECHOS and hidden Markov model toolkit (HTK) for validation. In an FSN-based task. ECHOS shows similar word accuracy while the recognition time is doubled because of object-oriented implementation. For a 8000-word continuous speech recognition task, using the lexical tree search algorithm different from the algorithm used in HTK, it increases the word error rate by $40\%$ relatively but reduces the recognition time to half.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.460-460
• /
• 2023

최근 범 지구적인 기후변화로 인해 집중호우가 빈번히 발생하고 침수피해가 증가하고 있다. 이에 따른 침수 피해 위험이 큰 지하상가, 지하 주차장, 반지하 주택 등의 침수 발생이 잦아지며 인명 및 재산 피해 발생이 커지고 있다. 이러한 지역은 인근 하수관로의 수위에 따라 침수 영향을 크게 받게 된다. 이에 따른 강우·유출 관계는 침수피해에 대해 대처하기 위해 시공간적 강우 특성이 반영된 하수관로 수위 예측이 중요하다고 판단된다. 이에 본 연구에서 수위 자료는 서울시 하수관로 수위 현황 자료를 활용하였으며, 강수량 자료는 서울 내 서초구 일대의 강수량 자료를 활용하여 연구를 진행하였다. 대상 지역은 저지대에 위치해 침수가 잦은 서초구 서초동으로 선정하였으며, 분석에 사용된 기간은 2012년부터 2021년까지의 수위 자료를 화용하여 이를 바탕으로 순환 신경망인 RNN의 일종이며, 다른 모델의 구조와 비교하여 더욱 간단하고 효율적인 ESN(Echo State Network) 알고리즘을 사용하여 수위 예측을 진행하였다. 분석을 위해 대상 지역의 강수 사상이 발생하여 하수관로의 수위의 변동이 큰 기간을 선정하여 분석을 실시하였다. 2012년부터 2018년까지의 자료를 학습(training) 자료로 활용하였으며, 모형의 검증 위해 통계분석을 실시하여 검증하였다.