• 한국어정보학회:학술대회논문집
• /
• 2016.10a
• /
• pp.333-336
• /
• 2016

본 연구에서는 개체명 인식의 성능을 향상시키기 위해, 가중 투표 방법을 이용하여 개체명 인식 모델을 앙상블 하는 방법을 제안한다. 각 모델은 Conditional Random Fields의 변형 알고리즘을 사용하여 학습하고, 모델들의 가중치는 다목적 함수 최적화 기법인 NSGA-II 알고리즘으로 학습한다. 실험 결과 제안 시스템은 $F_1Score$ 기준으로 87.62%의 성능을 보여, 단독 모델 중 가장 높은 성능을 보인 방법보다 2.15%p 성능이 향상되었다.

• Journal of Korea Multimedia Society
• /
• v.19 no.3
• /
• pp.531-538
• /
• 2016

In this paper, we propose a novel post-processing method of semantic segmentation from indoor scenes with RGBD inputs. For accurate segmentation, various post-processing methods such as superpixel from color edges or Conditional Random Field (CRF) method considering neighborhood connectivity have been used, but these methods are not efficient due to high complexity and computational cost. To solve this problem, we maximize the efficiency of post processing by using depth superpixel extracted from disparity image to handle object silhouette. Our experimental results show reasonable performances compared to previous methods in the post processing of semantic segmentation.

• Annual Conference on Human and Language Technology
• /
• 2012.10a
• /
• pp.12-15
• /
• 2012

본 논문은 한국어 형태소 분할 및 품사 태깅을 위해 조건부 랜덤 필드 (CRF: conditional random field)에 기반한 방식을 제안한다. 제안 방법은 1) 형태소 분할 단계 2) 품사 태깅 단계 3) 복합형태소 분할 및 태깅 단계의 세 단계로 이루어진다. 처음 두 단계는 CRF방법에 기반을 두고, 세 번째 단계에서는 일반화된 HMM (lattice-HMM)을 활용한다. 제안 방법은 세종 말뭉치 코퍼스에서 5-fold cross-validation로 평가한 결과, 약 96%의 품사 태깅 성능을 보여주었다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2008.06c
• /
• pp.539-544
• /
• 2008

이 논문에서 크기의 변화에 강인한 인공물 축출 알고리듬을 제안한다. 인공물은 크기 및 카메라 센서의 특성에 따라 영상에 다양한 크기로 나타난다. 이 논문은 이러한 크기 변화에 강인한 인공물 축출 방법을 제안한다. 우선 LoG(Laplacian of Gaussian)를 이용하여 최적의 크기를 찾아낸다. 이를 이용하여 우리는 이웃한 정보를 포함할 수 있는 MAP-MRF(Maximum A Posterior-Markov Random Field) 레이블링(Labeling) 방법을 기반으로 인공물 축출을 위한 비용함수를 제안하였다. 인공물은 서로 근처에 존재하기 때문이다. 여기서 정보 비용함수(Data cost function)는 방향 히스토그램(Orientation histogram)을 이용하여 정의하였고, 스무딩 비용함수(Smoothing cost function)는 ICM(Iterated Conditional Modes)을 이용하여 정의한다. 최종적으로 이 알고리듬을 위성영상에 적용하여 알고리듬의 성능을 증명한다.

• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.333-336
• /
• 2016

본 연구에서는 개체명 인식의 성능을 향상시키기 위해, 가중 투표 방법을 이용하여 개체명 인식 모델을 앙상블 하는 방법을 제안한다. 각 모델은 Conditional Random Fields의 변형 알고리즘을 사용하여 학습하고, 모델들의 가중치는 다목적 함수 최적화 기법인 NSGA-II 알고리즘으로 학습한다. 실험 결과 제안 시스템은 $F_1Score$기준으로 87.62%의 성능을 보여, 단독 모델 중 가장 높은 성능을 보인 방법보다 2.15%p 성능이 향상되었다.

• Annual Conference on Human and Language Technology
• /
• 2015.10a
• /
• pp.85-89
• /
• 2015

본 논문에서는 시퀀스 레이블링 문제에 적합하다고 알려진 Long Short Term Memory Recurrent Neural Network에 아웃풋간의 의존관계를 추가한 LSTM-CRF(Conditional Random Field)를 이용하여 생명과학분야 개체명 인식 시스템을 구축하였다. 학습 및 평가를 위해 BioNLP 2011-st REL data를 개체명 인식 실험에 사용하였으며, 실험결과 LSTM-CRF를 사용한 시스템은 81.83의 F1-score를 기록해, 기존의 시스템인 "BANNER"의 F1-score 81.96과 비슷한 성능을 보였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2018.10a
• /
• pp.613-616
• /
• 2018

Sequence tagging is an important task in Natural Language Processing (NLP), in which the Named Entity Recognition (NER) is the key issue. So far the most widely adopted model for NER in NLP is that of combining the neural network of bidirectional long short-term memory (BiLSTM) and the statistical sequence prediction method of Conditional Random Field (CRF). In this work, we improve the prediction accuracy of the BiLSTM by supporting an aligned word representation mechanism. We have performed experiments on multilingual (English, Spanish and Dutch) datasets and confirmed that our proposed model outperformed the existing state-of-the-art models.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.22 no.3
• /
• pp.303-309
• /
• 2004

In this paper, an adaptive Bayesian approach to image segmentation was developed for boundary detection. Both image intensities and texture information were used for obtaining better quality of the image segmentation by using the C programming language. Fuzzy c-mean clustering was applied fer the conditional probability density function, and Gibbs random field model was used for the prior probability density function. To simply test the algorithm, a synthetic image (256$\times$256) with a set of low gray values (50, 100, 150 and 200) was created and normalized between 0 and 1 n double precision. Results have been presented that demonstrate the effectiveness of the algorithm in segmenting the synthetic image, resulting in more than 99% accuracy when noise characteristics are correctly modeled. The algorithm was applied to the Antarctic mosaic that was generated using 1963 Declassified Intelligence Satellite Photographs. The accuracy of the resulting vector map was estimated about 300-m.

• International Journal of Computer Science & Network Security
• /
• v.22 no.1
• /
• pp.167-174
• /
• 2022

In metaheuristic algorithms such as Genetic Algorithm (GA), initial population has a significant impact as it affects the time such algorithm takes to obtain an optimal solution to the given problem. In addition, it may influence the quality of the solution obtained. In the machine learning field, feature selection is an important process to attaining a good performance model; Genetic algorithm has been utilized for this purpose by scientists. However, the characteristics of Genetic algorithm, namely random initial population generation from a vector of feature elements, may influence solution and execution time. In this paper, the use of a statistical algorithm has been introduced (Chi2) for feature relevant checks where p-values of conditional independence were considered. Features with low p-values were discarded and subject relevant subset of features to Genetic Algorithm. This is to gain a level of certainty of the fitness of features randomly selected. An ensembled-based learning model for Hepatitis has been developed for Hepatitis C stage classification. 1385 samples were used using Egyptian-dataset obtained from UCI repository. The comparative evaluation confirms decreased in execution time and an increase in model performance accuracy from 56% to 63%.

• ETRI Journal
• /
• v.44 no.5
• /
• pp.794-804
• /
• 2022

Transformer models have had a great impact on natural language processing (NLP) in recent years by realizing outstanding and efficient contextualized language models. Recent studies have used transformer-based language models for various NLP tasks, including Persian named entity recognition (NER). However, in complex tasks, for example, NER, it is difficult to determine which contextualized embedding will produce the best representation for the tasks. Considering the lack of comparative studies to investigate the use of different contextualized pretrained models with sequence modeling classifiers, we conducted a comparative study about using different classifiers and embedding models. In this paper, we use different transformer-based language models tuned with different classifiers, and we evaluate these models on the Persian NER task. We perform a comparative analysis to assess the impact of text representation and text classification methods on Persian NER performance. We train and evaluate the models on three different Persian NER datasets, that is, MoNa, Peyma, and Arman. Experimental results demonstrate that XLM-R with a linear layer and conditional random field (CRF) layer exhibited the best performance. This model achieved phrase-based F-measures of 70.04, 86.37, and 79.25 and word-based F scores of 78, 84.02, and 89.73 on the MoNa, Peyma, and Arman datasets, respectively. These results represent state-of-the-art performance on the Persian NER task.