• Journal of Multimedia Information System
• /
• v.9 no.2
• /
• pp.93-102
• /
• 2022

Dependency parsing is a decision problem of the syntactic relation between words in a sentence. Recently, deep learning models are used for dependency parsing based on the word representations in a continuous vector space. However, it causes a mislabeled tagging problem for the proper nouns that rarely appear in the training corpus because it is difficult to express out-of-vocabulary (OOV) words in a continuous vector space. To solve the OOV problem in dependency parsing, we explored the proper noun embedding method according to the embedding unit. Before representing words in a continuous vector space, we replace the proper nouns with a special token and train them for the contextual features by using the multi-layer bidirectional LSTM. Two models of the syllable-based and morpheme-based unit are proposed for proper noun embedding and the performance of the dependency parsing is more improved in the ensemble model than each syllable and morpheme embedding model. The experimental results showed that our ensemble model improved 1.69%p in UAS and 2.17%p in LAS than the same arc-eager approach-based Malt parser.

• Journal of the Institute of Convergence Signal Processing
• /
• v.11 no.2
• /
• pp.137-142
• /
• 2010

Although multi-model speech recognizer has been shown to be quite successful in noisy speech recognition, the results were based on general speech front-ends which do not take into account noise adaptation techniques. In this paper, for the accurate evaluation of the multi-model based speech recognizer, we adopted a quite noise-robust speech front-end, AFE, which was proposed by the ETSI for the noisy DSR environment. For the performance comparison, the MTR which is known to give good results in the DSR environment has been used. Also, we modified the structure of the multi-model based speech recognizer to improve the recognition performance. N reference HMMs which are most similar to the input noisy speech are used as the acoustic models for recognition to cope with the errors in the selection of the reference HMMs and the noise signal variability. In addition, multiple SNR levels are used to train each of the reference HMMs to improve the robustness of the acoustic models. From the experimental results on the Aurora 2 databases, we could see better recognition rates using the modified multi-model based speech recognizer compared with the previous method.

• Journal of Korea Water Resources Association
• /
• v.51 no.7
• /
• pp.607-616
• /
• 2018

Flow regime-based ANFIS Dam Inflow Prediction (FADIP) model is developed and compared with ANFIS Dam Inflow Prediction (ADIP) model in this study. The selected study area is the Chungju and Soyang multi-purpose dam watersheds in South Korea. The dam inflow, precipitation and monthly weather forecast information are used as input variables of the models. The training and validation periods of the models are 1987~2010 for Chungju and 1984~2010 for Soyang dam watershed. The testing periods for both watersheds are 2011~2016. The results of training and validation indicate that FADIP has better training ability than ADIP for predicting dam inflow in normal and low flow regimes. In the result of testing, ADIP shows low predictability of dam inflow in the low flow regime due to the model tuning on all flow regime together. However, FADIP demonstrates the improved accuracy over the entire period compared to ADIP, especially during the normal and low flow seasons. It is concluded that FADIP is valuable for the prediction of dam inflow in the case of drought years, and useful for water supply management of the multi-purpose dam.

• Journal of the Korean Geotechnical Society
• /
• v.20 no.8
• /
• pp.77-87
• /
• 2004

In this paper a back-propagation neural network model is developed to estimate the preconsolidation pressure of Korean soft soils based on 176 oedometer tests and 63 piezocone test results, which were compiled from 11 sites - western and southern parts of Korea. Only 147 data were used for the training of the neural network and 29 data, which were not used during the training phase, were used for the verification of trained network. Empirical and theoretical models were compared with the developed neural network model. A simple 4-4-9-1 multi-layered neural network has been developed. The cone tip resistance $q_T$ penetration pore pressure $u_2$, total overburden pressure $\sigma_{vo}$ and effective overburden pressure $\sigma'_{vo}$ were selected as input variables. The developed neural network model was validated by comparing the prediction results of the proposed neural network model for the new data which were not used for the training of the model with the measured preconsolidation pressures. It can also predict more precise and reliable preconsolidation pressures than the analytical and empirical model. Furthermore, it can be carefully concluded that neural network model can be used as a generalized model for prediction of preconsolidation pressure throughout Korea since developed model shows good performance for the new data which were not used in both training and testing data.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.149.5-149
• /
• 2001

The robot soccer simulation game is a dynamic multi-agent environment. In this paper we suggest a new reinforcement learning approach to each agent´s dynamic positioning in such dynamic environment. Reinforcement learning is the machine learning in which an agent learns from indirect, delayed reward an optimal policy to choose sequences of actions that produce the greatest cumulative reward. Therefore the reinforcement learning is different from supervised learning in the sense that there is no presentation of input-output pairs as training examples. Furthermore, model-free reinforcement learning algorithms like Q-learning do not require defining or learning any models of the surrounding environment. Nevertheless ...

• Proceedings of the Optical Society of Korea Conference
• /
• 1991.06a
• /
• pp.35-40
• /
• 1991

In this paper, a new adaptive learning algorithm, Training by Adaptive Gain (TAG) for optical implementation of large-sized neural networks has been developed and its electro-optical implementation for 2-dimensional input and output neurons has been demostrated. The 4-dimensional global fixed interconnections and 2-dimensional adaptive gain-controls are implemented by multi-facet computer generated holograms and LCTV spatial light modulators, respectively. When the input signals pass through optical system to the output classifying layer, the TAG adaptive learning algorithm is implemented by a personal computer. The system classifies three 5$\times$5 input patterns correctly.

• International Journal of Computer Science & Network Security
• /
• v.23 no.9
• /
• pp.129-133
• /
• 2023

Classification or prediction problem is how to solve it using a specific feature to obtain the predicted class. A wheat seeds specifications 4 3 classes of seeds will be used in a prediction process. A multi linear regression will be built, and a prediction error ratio will be calculated. To enhance the prediction ratio an ANN model will be built and trained. The obtained results will be examined to show how to make a prediction tool capable to compute a predicted class number very close to the target class number.

• Journal of IKEEE
• /
• v.20 no.1
• /
• pp.94-102
• /
• 2016

This paper proposes a new prediction method to reduce times and labor of repetitive multi-physics simulation. To achieve exact results from the whole simulation processes, complex modeling and huge amounts of time are required. Current multi-physics analysis focuses on the simulation method itself and the simulation environment to reduce times and labor. However this paper proposes an alternative way to reduce simulation times and labor by exploiting machine learning algorithm trained with data set from simulation results. Through comparing each machine learning algorithm, Gaussian Process Regression showed the best performance with under 100 training data and how similar results can be achieved through machine-learning without a complex simulation process. Given trained machine learning algorithm, it's possible to predict the result after changing some features of the simulation model just in a few second. This new method will be helpful to effectively reduce simulation times and labor because it can predict the results before more simulation.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.2
• /
• pp.480-502
• /
• 2022

This work assesses the degree of satisfaction tourists receive as final recipients in a tourism destination based on the fact that satisfied tourists can make a significant contribution to the growth and continuous improvement of a tourism business. The work considers Pokhara, the tourism capital of Nepal as a prefecture of study. A stratified sampling methodology with open-ended survey questions is used as a primary source of data for a sample size of 1019 for both international and domestic tourists. The data collected through a survey is processed using a data mining tool to perform multi-dimensional analysis to discover information patterns and visualize clusters. Further, supervised machine learning algorithms, kNN, Decision tree, Support vector machine, Random forest, Neural network, Naive Bayes, and Gradient boost are used to develop models for training and prediction purposes for the survey data. To find the best model for prediction purposes, different performance matrices are used to evaluate a model for performance, accuracy, and robustness. The best model is used in constructing a learning-enabled model for predicting tourists as satisfied, neutral, and unsatisfied visitors. This work is very important for tourism business personnel, government agencies, and tourism stakeholders to find information on tourist satisfaction and factors that influence it. Though this work was carried out for Pokhara city of Nepal, the study is equally relevant to any other tourism destination of similar nature.

• ETRI Journal
• /
• v.39 no.3
• /
• pp.398-405
• /
• 2017

In this paper, an artificial stereo extension method that creates stereophonic sound from a mono sound source is proposed. The proposed method first trains deep neural networks (DNNs) that model the nonlinear relationship between the dominant and residual signals of the stereo channel. In the training stage, the band-wise log spectral magnitude and unwrapped phase of both the dominant and residual signals are utilized to model the nonlinearities of each sub-band through deep architecture. From that point, stereo extension is conducted by estimating the residual signal that corresponds to the input mono channel signal with the trained DNN model in a sub-band domain. The performance of the proposed method was evaluated using a log spectral distortion (LSD) measure and multiple stimuli with a hidden reference and anchor (MUSHRA) test. The results showed that the proposed method provided a lower LSD and higher MUSHRA score than conventional methods that use hidden Markov models and DNN with full-band processing.