• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.34-36
• /
• 2019

기존 딥 러닝을 이용한 형태소 분석 및 품사 태깅(Part-Of-Speech tagging)은 feed-forward neural network에 CRF를 결합하는 방법이나 sequence-to-sequence 모델을 이용한 방법 등의 다양한 모델들이 연구되었다. 본 논문에서는 한국어 형태소 분석 및 품사 태깅을 수행하기 위하여 최근 자연어처리 태스크에서 많은 성능 향상을 보이고 있는 BERT를 기반으로 한 음절 단위 LSTM-CRF 모델을 제안한다. BERT는 양방향성을 가진 트랜스포머(transformer) 인코더를 기반으로 언어 모델을 사전 학습한 것이며, 본 논문에서는 한국어 대용량 코퍼스를 어절 단위로 사전 학습한 KorBERT를 사용한다. 실험 결과, 본 논문에서 제안한 모델이 기존 한국어 형태소 분석 및 품사 태깅 연구들 보다 좋은 (세종 코퍼스) F1 98.74%의 성능을 보였다.

• The Bulletin of The Korean Astronomical Society
• /
• v.46 no.2
• /
• pp.45.3-46
• /
• 2021

Supernovae classes have been defined phenomenologically, based on spectral features and time series data, since the specific details of the physics of the different explosions remain unrevealed. However, the number of these classes is increasing as objects with new features are observed, and the next generation of large-surveys will only bring more variety to our attention. We apply the machine learning technique of multi-label classification to the spectra of supernovae. By measuring the probabilities of specific features or 'tags' in the supernova spectra, we can compress the information from a specific object down to that suitable for a human or database scan, without the need to directly assign to a reductive 'class'. We use logistic regression to assign tag probabilities, and then a feed-forward neural network to filter the objects into the standard set of classes, based solely on the tag probabilities. We present STag, a software package that can compute these tag probabilities and make spectral classifications.

• Structural Engineering and Mechanics
• /
• v.71 no.5
• /
• pp.545-552
• /
• 2019

The analysis and design of complex structures like sandwich-panel elements are difficult; the use of finite element method for the analysis is complicated and time consuming when non-linear effects are considered. On the other hand, artificial neural network (ANN) models can capture the non-linear effects and its application requires lesser computational demand. Two ANN models were trained, tested and validated to compute the force for a given displacement of a sandwich-type roof element; 2555 force and element deformation pairs were used for training the ANN models. For the models trained without considering the damping effect, there were two values in the input layer: maximum displacement and current displacement, and for the model considering damping, displacement from the previous step was used as an additional input. Totally, 400 ANN models were trained. Results show that there is a good agreement between the experimental and simulated data, and the models showed a good performance with a mean square error value of 4548.85. Both the ANN models could simulate the inelastic behaviour, loss of rigidity, and evolution of permanent displacements. The models could also interpolate and extrapolate, which enables them to be used as an analysis and design tool for such complex elements.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.23 no.12
• /
• pp.154-161
• /
• 2009

A fuel cell power plant is a very complex system which has various control loops with some non-linearity. For control of a fuel cell power plant, dynamic models of fuel cell stacks have been developed and simplified process flow diagrams of a fuel cell power plant has been presented. Using such a model of a Molten Carbonate Fuel Cell (MCFC) power plant, this paper deals with development of an intelligent setpoint reference governor (I-SRG) to find the optimal setpoints and feed forward control inputs for the plant power demand. The I-SRG is implemented with neural network by using Particle Swarm Optimization (PSO) algorithm based on system constraints and performance objectives. The feasibility of the I-SRG is shown through simulation of an MCFC power plant for tracking control of its power demand.

• Annual Conference on Human and Language Technology
• /
• 1989.10a
• /
• pp.137-141
• /
• 1989

본 논문에서는 신경망 구조의 한 모델인 feed-forward multi-layered network에 역전파 학습(back-propagation learning) 기법을 이용하여 필기체 한글 숫자를 인식하고 그 가능성을 보였다. 문자 인식에 있어 입력 대상의 모양이 왜곡되거나, 대상의 크기 혹은 위치의 변화 등과 같은 잡음 (noise)에 대해서 정확히 대상을 인식하는 데는 대상의 구조 추출에 크게 관여되므로 한글의 구조 추출에 적합하다고 생각되는 bar mask 투사법을 제안하였다. 모델의 학습을 필기체 한글 숫자 16자의 입력 패턴과 타겟 ( target) 입력의 쌍을 이용해 학습시켰다. 또한, 모델의 인식 정도를 측정해 보기 위해 시험패턴을 적용하여 훈련된 패턴과 훈련되지 않은 패턴간의 인식률을 비교하여 보았다.

• Journal of Information Processing Systems
• /
• v.10 no.3
• /
• pp.443-458
• /
• 2014

An extreme learning machine (ELM) is a recently proposed learning algorithm for a single-layer feed forward neural network. In this paper we studied the ensemble of ELM by using a bagging algorithm for facial expression recognition (FER). Facial expression analysis is widely used in the behavior interpretation of emotions, for cognitive science, and social interactions. This paper presents a method for FER based on the histogram of orientation gradient (HOG) features using an ELM ensemble. First, the HOG features were extracted from the face image by dividing it into a number of small cells. A bagging algorithm was then used to construct many different bags of training data and each of them was trained by using separate ELMs. To recognize the expression of the input face image, HOG features were fed to each trained ELM and the results were combined by using a majority voting scheme. The ELM ensemble using bagging improves the generalized capability of the network significantly. The two available datasets (JAFFE and CK+) of facial expressions were used to evaluate the performance of the proposed classification system. Even the performance of individual ELM was smaller and the ELM ensemble using a bagging algorithm improved the recognition performance significantly.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2007.06a
• /
• pp.171-174
• /
• 2007

A realization for FF-PID(Feed-Forward PID) controlling system with backward propagation algorithm and image pattern recognition is presented in this paper. The pattern recognition used backward propagation of nervous network is teaming. FF-PID is enhanced the response characteristic of moving image by using the controlling value which is output error for the target value of nervous system. In conclusion of experiment, the system is shown that the response is worked as 2.7sec that is enhanced round 15% in comparison with general difference image algorithm. The system is able to control a moving object with effect.

• Journal of Biomedical Engineering Research
• /
• v.19 no.5
• /
• pp.455-468
• /
• 1998

For the objective interpretation of cerebral metabolic patterns in epilepsy patients, we developed computer-aided classifier using artificial neural network. We studied interictal brain FDG PET scans of 257 epilepsy patients who were diagnosed as normal(n=64), L TLE (n=112), or R TLE (n=81) by visual interpretation. Automatically segmented volume of interest (VOI) was used to reliably extract the features representing patterns of cerebral metabolism. All images were spatially normalized to MNI standard PET template and smoothed with 16mm FWHM Gaussian kernel using SPM96. Mean count in cerebral region was normalized. The VOls for 34 cerebral regions were previously defined on the standard template and 17 different counts of mirrored regions to hemispheric midline were extracted from spatially normalized images. A three-layer feed-forward error back-propagation neural network classifier with 7 input nodes and 3 output nodes was used. The network was trained to interpret metabolic patterns and produce identical diagnoses with those of expert viewers. The performance of the neural network was optimized by testing with 5~40 nodes in hidden layer. Randomly selected 40 images from each group were used to train the network and the remainders were used to test the learned network. The optimized neural network gave a maximum agreement rate of 80.3% with expert viewers. It used 20 hidden nodes and was trained for 1508 epochs. Also, neural network gave agreement rates of 75~80% with 10 or 30 nodes in hidden layer. We conclude that artificial neural network performed as well as human experts and could be potentially useful as clinical decision support tool for the localization of epileptogenic zones.

• Steel and Composite Structures
• /
• v.27 no.4
• /
• pp.427-438
• /
• 2018

In this paper, the single and multi-objective optimization of thin-walled conical tubes with different types of indentations under axial impact has been investigated using surrogate models called metamodels. The geometry of tapered thin-walled tubes has been studied in order to achieve maximum specific energy absorption (SEA) and minimum peak crushing force (PCF). The height, radius, thickness, tapered angle of the tube, and the radius of indentation have been considered as design variables. Based on the design of experiments (DOE) method, the generated sample points are computed using the explicit finite element code. Different surrogate models including Kriging, Feed Forward Neural Network (FNN), Radial Basis Neural Network (RNN), and Response Surface Modelling (RSM) comprised to evaluate the appropriation of such models. The comparison study between surrogate models and the exploration of indentation shapes have been provided. The obtained results show that the RNN method has the minimum mean squared error (MSE) in training points compared to the other methods. Meanwhile, optimization based on surrogate models with lower values of MSE does not provide optimum results. The RNN method demonstrates a lower crashworthiness performance (with a lower value of 125.7% for SEA and a higher value of 56.8% for PCF) in comparison to RSM with an error order of $10^{-3}$. The SEA values can be increased by 17.6% and PCF values can be decreased by 24.63% by different types of indentation. In a specific geometry, higher SEA and lower PCF require triangular and circular shapes of indentation, respectively.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.520-520
• /
• 2023

제주도 동부 중산간 지역은 화산암으로 구성된 지하지질로 인해 지하수위의 변동폭이 크고 변동양상이 복잡하여 인공신경망(Artificial Neural Network, ANN) 모델 등을 활용한 지하수위의 예측이 어렵다. ANN에 적용되는 활성화함수에 따라 지하수의 예측성능은 달라질 수 있으므로 활성화함수의 비교분석 후 적절한 활성화함수의 사용이 반드시 필요하다. 본 연구에서는 5개 활성화함수(sigmoid, hyperbolic tangent(tanh), Rectified Linear Unit(ReLU), Leaky Rectified Linear Unit(Leaky ReLU), Exponential Linear Unit(ELU))를 제주도 동부 중산간지역에 위치한 2개 지하수 관정에 대해 비교분석하여 최적 활성화함수 도출을 목표로 한다. 또한 최적 활성화함수를 활용한 ANN의 적용성을 평가하기 위해 최근 널리 사용되고 있는 순환신경망 모델인 Long Short-Term Memory(LSTM) 모델과 비교분석 하였다. 그 결과, 2개 관정 중 지하수위 변동폭이 상대적으로 큰 관정은 ELU 함수, 상대적으로 작은 관정은 Leaky ReLU 함수가 지하수위 예측에 적절하였다. 예측성능이 가장 낮은 활성화함수는 sigmoid 함수로 나타나 첨두 및 최저 지하수위 예측 시 사용을 지양해야 할 것으로 판단된다. 도출된 최적 활성화함수를 사용한 ANN-ELU 모델 및 ANN-Leaky ReLU 모델을 LSTM 모델과 비교분석한 결과 대등한 지하수위 예측성능을 나타내었다. 이것은 feed-forward 방식인 ANN 모델을 사용하더라도 적절한 활성화함수를 사용하면 최신 순환신경망과 대등한 결과를 도출하여 활용 가능성이 충분히 있다는 것을 의미한다. 마지막으로 LSTM 모델은 가장 적절한 예측성능을 나타내어 다양한 인공지능 모델의 예측성능 비교를 위한 기준이 되는 참고모델로 활용 가능하다. 본 연구에서 제시한 방법은 지하수위 예측과 더불어 하천수위 예측 등 다양한 시계열예측 및 분석연구에 유용하게 사용될 수 있다.