• Journal of Information Processing Systems
• /
• 제16권4호
• /
• pp.809-819
• /
• 2020

Gender prediction accuracy increases as convolutional neural network (CNN) architecture evolves. This paper compares voting and ensemble schemes to utilize the already trained five CNN models to further improve gender prediction accuracy. The majority voting usually requires odd-numbered models while the proposed softmax-based voting can utilize any number of models to improve accuracy. The ensemble of CNN models combined with one more fully-connected layer requires further tuning or training of the models combined. With experiments, it is observed that the voting or ensemble of CNN models leads to further improvement of gender prediction accuracy and that especially softmax-based voters always show better gender prediction accuracy than majority voters. Also, compared with softmax-based voters, ensemble models show a slightly better or similar accuracy with added training of the combined CNN models. Softmax-based voting can be a fast and efficient way to get better accuracy without further training since the selection of the top accuracy models among available CNN pre-trained models usually leads to similar accuracy to that of the corresponding ensemble models.

• 한국컴퓨터정보학회논문지
• /
• 제27권2호
• /
• pp.15-23
• /
• 2022

특허 분야에서 자연어처리(Natural Language Processing) 태스크는 특허문헌의 언어적 특이성으로 문제 해결의 난이도가 높은 과제임에 따라 한국 특허문헌에 최적화된 언어모델의 연구가 시급한 실정이다. 최근 자연어처리 분야에서는 특정 도메인에 특화되게 사전 학습(Pre-trained)한 언어모델을 구축하여 관련 분야의 다양한 태스크에서 성능을 향상시키려는 시도가 지속적으로 이루어지고 있다. 그 중, ELECTRA는 Google이 BERT 이후에 RTD(Replaced Token Detection)라는 새로운 방식을 제안하며 학습 효율성을 높인 사전학습 언어모델이다. 본 연구에서는 대량의 한국 특허문헌 데이터를 사전 학습한 KorPatELECTRA를 제안한다. 또한, 특허 문헌의 특성에 맞게 학습 코퍼스를 정제하고 특허 사용자 사전 및 전용 토크나이저를 적용하여 최적화된 사전 학습을 진행하였다. KorPatELECTRA의 성능 확인을 위해 실제 특허데이터를 활용한 NER(Named Entity Recognition), MRC(Machine Reading Comprehension), 특허문서 분류 태스크를 실험하였고 비교 대상인 범용 모델에 비해 3가지 태스크 모두에서 가장 우수한 성능을 확인하였다.

• Journal of Information Processing Systems
• /
• 제19권5호
• /
• pp.673-687
• /
• 2023

Convolutional neural network-based deep learning technology is the most commonly used in image identification, but it requires large-scale data for training. Therefore, application in specific fields in which data acquisition is limited, such as in the military, may be challenging. In particular, the identification of ground weapon systems is a very important mission, and high identification accuracy is required. Accordingly, various studies have been conducted to achieve high performance using small-scale data. Among them, the ensemble method, which achieves excellent performance through the prediction average of the pre-trained models, is the most representative method; however, it requires considerable time and effort to find the optimal combination of ensemble models. In addition, there is a performance limitation in the prediction results obtained by using an ensemble method. Furthermore, it is difficult to obtain the ensemble effect using models with imbalanced classification accuracies. In this paper, we propose a transfer learning-based feature fusion technique for heterogeneous models that extracts and fuses features of pre-trained heterogeneous models and finally, fine-tunes hyperparameters of the fully connected layer to improve the classification accuracy. The experimental results of this study indicate that it is possible to overcome the limitations of the existing ensemble methods by improving the classification accuracy through feature fusion between heterogeneous models based on transfer learning.

• 한국컴퓨터정보학회논문지
• /
• 제25권5호
• /
• pp.187-197
• /
• 2020

최근 신경망 기반의 학습 알고리즘인 딥 러닝 기술의 발전으로 인해 텍스트의 문맥을 고려한 문서 임베딩 모델이 다양하게 고안되었으며, 특히 대량의 텍스트 데이터를 사용하여 학습을 수행한 사전 학습 언어 모델을 사용하여 분석 문서의 벡터를 추론하는 방식의 임베딩이 활발하게 연구되고 있다. 하지만 기존의 사전 학습 언어 모델을 사용하여 새로운 텍스트에 대한 임베딩을 수행할 경우 해당 텍스트가 가진 고유한 정보를 충분히 활용하지 못한다는 한계를 가지며, 이는 특히 텍스트가 가진 토큰의 수에 큰 영향을 받는 것으로 알려져 있다. 이에 본 연구에서는 다수의 토큰을 포함한 장문 텍스트의 정보를 최대한 활용하여 해당 텍스트의 벡터를 도출할 수 있는 자기 지도 학습 기반의 사전 학습 언어 모델 미세 조정 방법을 제안한다. 또한, 제안 방법론을 실제 뉴스 기사에 적용하여 문서 벡터를 도출하고 이를 활용하여 뉴스의 카테고리 분류 실험을 수행하는 외부적인 임베딩 평가를 수행함으로써, 제안 방법론과 기존 문서 임베딩 모델과의 성능을 평가하였다. 그 결과 제안 방법론을 통해 도출된 벡터가 텍스트의 고유 정보를 충분히 활용함으로써, 문서의 특성을 더욱 정확하게 표현할 수 있음을 확인하였다.

• International Journal of Computer Science & Network Security
• /
• 제21권9호
• /
• pp.223-229
• /
• 2021

Classification of different blood cell types is an essential task for human's medical treatment. The white blood cells have different types of cells. Counting total White Blood Cells (WBC) and differential of the WBC types are required by the physicians to diagnose the disease correctly. This paper used transfer learning methods to the pre-trained deep learning models to classify different WBCs. The best pre-trained model was Inception ResNetV2 with Adam optimizer that produced classification accuracy of 98.4% for the dataset comprising four types of WBCs.

• International Journal of Computer Science & Network Security
• /
• 제21권8호
• /
• pp.65-70
• /
• 2021

Waste accumulation is becoming a significant challenge in most urban areas and if it continues unchecked, is poised to have severe repercussions on our environment and health. The massive industrialisation in our cities has been followed by a commensurate waste creation that has become a bottleneck for even waste management systems. While recycling is a viable solution for waste management, it can be daunting to classify waste material for recycling accurately. In this study, transfer learning models were proposed to automatically classify wastes based on six materials (cardboard, glass, metal, paper, plastic, and trash). The tested pre-trained models were ResNet50, VGG16, InceptionV3, and Xception. Data augmentation was done using a Generative Adversarial Network (GAN) with various image generation percentages. It was found that models based on Xception and VGG16 were more robust. In contrast, models based on ResNet50 and InceptionV3 were sensitive to the added machine-generated images as the accuracy degrades significantly compared to training with no artificial data.

• 한국멀티미디어학회논문지
• /
• 제21권2호
• /
• pp.98-107
• /
• 2018

In this paper, we propose a method to classify the skeletal maturity with a small amount of hand wrist X-ray image using deep learning-based meta-learning. General deep-learning techniques require large amounts of data, but in many cases, these data sets are not available for practical application. Lack of learning data is usually solved through transfer learning using pre-trained models with large data sets. However, transfer learning performance may be degraded due to over fitting for unknown new task with small data, which results in poor generalization capability. In addition, medical images require high cost resources such as a professional manpower and mcuh time to obtain labeled data. Therefore, in this paper, we use meta-learning that can classify using only a small amount of new data by pre-trained models trained with various learning tasks. First, we train the meta-model by using a separate data set composed of various learning tasks. The network learns to classify the bone maturity using the bone maturity data composed of the radiographs of the wrist. Then, we compare the results of the classification using the conventional learning algorithm with the results of the meta learning by the same number of learning data sets.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권6호
• /
• pp.1692-1705
• /
• 2024

In this paper, we present a method that integrates a Grammar Transducer as an external language model to enhance the accuracy of the pre-trained Korean End-to-end (E2E) Automatic Speech Recognition (ASR) model. The E2E ASR model utilizes the Connectionist Temporal Classification (CTC) loss function to derive hypothesis sentences from input audio. However, this method reveals a limitation inherent in the CTC approach, as it fails to capture language information from transcript data directly. To overcome this limitation, we propose a fusion approach that combines a clause-level n-gram language model, transformed into a Weighted Finite-State Transducer (WFST), with the E2E ASR model. This approach enhances the model's accuracy and allows for domain adaptation using just additional text data, avoiding the need for further intensive training of the extensive pre-trained ASR model. This is particularly advantageous for Korean, characterized as a low-resource language, which confronts a significant challenge due to limited resources of speech data and available ASR models. Initially, we validate the efficacy of training the n-gram model at the clause-level by contrasting its inference accuracy with that of the E2E ASR model when merged with language models trained on smaller lexical units. We then demonstrate that our approach achieves enhanced domain adaptation accuracy compared to Shallow Fusion, a previously devised method for merging an external language model with an E2E ASR model without necessitating additional training.

• Korean Journal of Radiology
• /
• 제22권12호
• /
• pp.2073-2081
• /
• 2021

Deep learning-based applications have great potential to enhance the quality of medical services. The power of deep learning depends on open databases and innovation. Radiologists can act as important mediators between deep learning and medicine by simultaneously playing pioneering and gatekeeping roles. The application of deep learning technology in medicine is sometimes restricted by ethical or legal issues, including patient privacy and confidentiality, data ownership, and limitations in patient agreement. In this paper, we present an open platform, MI2RLNet, for sharing source code and various pre-trained weights for models to use in downstream tasks, including education, application, and transfer learning, to encourage deep learning research in radiology. In addition, we describe how to use this open platform in the GitHub environment. Our source code and models may contribute to further deep learning research in radiology, which may facilitate applications in medicine and healthcare, especially in medical imaging, in the near future. All code is available at https://github.com/mi2rl/MI2RLNet.

• 한국산업정보학회논문지
• /
• 제29권3호
• /
• pp.79-87
• /
• 2024

최근 텍스트분류, 감성분석, 질의응답 등의 자연어 처리를 위해서 사전학습언어모델(Pre-trained Language Model, PLM)의 중요성은 날로 강조되고 있다. 한국어 PLM은 범용적인 도메인의 자연어 처리에서 높은 성능을 보이나 금융, 제조, 법률, 의료 등의 특화된 도메인에서는 성능이 미약하다. 본 연구는 금융도메인 뿐만 아니라 범용도메인에서도 우수한 성능을 보이는 금융특화 언어모델의 구축을 위해 언어모델의 학습과정과 미세조정 방법을 제안하는 것이 주요 목표이다. 금융도메인 특화언어모델을 구축하는 과정은 (1) 금융데이터 수집 및 전처리, (2) PLM 또는 파운데이션 모델 등 모델 아키텍처 선정, (3) 도메인 데이터 학습과 인스트럭션 튜닝, (4) 모델 검증 및 평가, (5) 모델 배포 및 활용 등으로 구성된다. 이를 통해 금융도메인의 특성을 살린 사전학습 데이터 구축방안과 효율적인 LLM 훈련방법인 적응학습과 인스트럭션 튜닝기법을 제안하였다.