• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.234-238
• /
• 2023

미세 조정은 대부분의 연구에서 사전학습 모델을 위한 표준 기법으로 활용되고 있으나, 최근 초거대 모델의 등장과 환경 오염 등의 문제로 인해 더 효율적인 사전학습 모델 활용 방법이 요구되고 있다. 패턴 추출 학습은 사전학습 모델을 효율적으로 활용하기 위해 제안된 방법으로, 본 논문에서는 한국어 주장 탐지 및 입장 분류를 위해 패턴 추출 학습을 활용하는 모델을 구현하였다. 우리는 기존 미세 조정 방식 모델과의 비교 실험을 통해 본 논문에서 구현한 한국어 주장 탐지 및 입장 분류 모델이 사전학습 단계에서 학습한 모델의 내부 지식을 효과적으로 활용할 수 있음을 보였다.

• Journal of Information Processing Systems
• /
• v.8 no.4
• /
• pp.575-588
• /
• 2012

Exploiting unlabeled text data with a relatively small labeled corpus has been an active and challenging research topic in text mining, due to the recent growth of the amount of biomedical literature. Biomedical named-entity recognition is an essential prerequisite task before effective text mining of biomedical literature can begin. This paper proposes an Active Co-Training (ACT) algorithm for biomedical named-entity recognition. ACT is a semi-supervised learning method in which two classifiers based on two different feature sets iteratively learn from informative examples that have been queried from the unlabeled data. We design a new classification problem to measure the informativeness of an example in unlabeled data. In this classification problem, the examples are classified based on a joint view of a feature set to be informative/non-informative to both classifiers. To form the training data for the classification problem, we adopt a query-by-committee method. Therefore, in the ACT, both classifiers are considered to be one committee, which is used on the labeled data to give the informativeness label to each example. The ACT method outperforms the traditional co-training algorithm in terms of f-measure as well as the number of training iterations performed to build a good classification model. The proposed method tends to efficiently exploit a large amount of unlabeled data by selecting a small number of examples having not only useful information but also a comprehensive pattern.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.3
• /
• pp.57-65
• /
• 1995

This paper presents a new classified vector quantization (VQ) technique using a neural network model in the transform domain. Prior to designing a codebook, the proposed approach extracts class features from a set of images using self-organizing feature map (SOFM) that has the pattern recognition characteristics and the same as VQ objective. Since we extract the class features from the training images unlike previous approaches, the reconstructed image quality is improved. Moreover, exploiting the adaptivity of the neural network model makes our approach be easily applied to designing a new vector quantizer when the processed image characteristics are changed. After the generalized BFOS algorithm allocates the given bits to each class, codebooks of each class are also generated using SOFM for the maximal reconstructed image quality. In experimental results using monochromatic images, we obtained a good visual quality in the reconstructed image. Also, PSNR is comparable to that of other classified VQ technique and is higher than that of JPEG baseline system.