• ETRI Journal
• /
• 제39권5호
• /
• pp.643-651
• /
• 2017

Human age estimation is one of the key factors in the field of Human-Robot Interaction/Human-Computer Interaction (HRI/HCI). Owing to the development of deep-learning technologies, age recognition has recently been attempted. In general, however, deep learning techniques require a large-scale database, and for age learning with variations, a conventional database is insufficient. For this reason, we propose an age estimation method using artificially generated data. Image data are artificially generated through 3D information, thus solving the problem of shortage of training data, and helping with the training of the deep-learning technique. Augmentation using 3D has advantages over 2D because it creates new images with more information. We use a deep architecture as a pre-trained model, and improve the estimation capacity using artificially augmented training images. The deep architecture can outperform traditional estimation methods, and the improved method showed increased reliability. We have achieved state-of-the-art performance using the proposed method in the Morph-II dataset and have proven that the proposed method can be used effectively using the Adience dataset.

• Journal of information and communication convergence engineering
• /
• 제20권3호
• /
• pp.219-225
• /
• 2022

In this study, we propose training a support vector machine (SVM) model on top of deep networks for detecting Covid-19 from chest X-ray images. We started by gathering a real chest X-ray image dataset, including positive Covid-19, normal cases, and other lung diseases not caused by Covid-19. Instead of training deep networks from scratch, we fine-tuned recent pre-trained deep network models, such as DenseNet121, MobileNet v2, Inception v3, Xception, ResNet50, VGG16, and VGG19, to classify chest X-ray images into one of three classes (Covid-19, normal, and other lung). We propose training an SVM model on top of deep networks to perform a nonlinear combination of deep network outputs, improving classification over any single deep network. The empirical test results on the real chest X-ray image dataset show that deep network models, with an exception of ResNet50 with 82.44%, provide an accuracy of at least 92% on the test set. The proposed SVM on top of the deep network achieved the highest accuracy of 96.16%.

• 지능정보연구
• /
• 제24권3호
• /
• pp.1-19
• /
• 2018

최근 딥러닝은 오디오, 텍스트 및 이미지 데이터와 같은 비 체계적인 데이터를 대상으로 다양한 추정, 분류 및 예측 문제에 사용 및 적용되고 있다. 특히, 의류산업에 적용될 경우 딥러닝 기법을 활용한 의류 인식, 의류 검색, 자동 제품 추천 등의 심층 학습을 기반으로 한 응용이 가능하다. 이 때의 핵심모형은 합성곱 신경망을 사용한 이미지 분류이다. 합성곱 신경망은 입력이 전달되고 출력에 도달하는 과정에서 가중치와 같은 매개 변수를 학습하는 뉴런으로 구성되고, 영상 분류에 가장 적합한 방법론으로 사용된다. 기존의 의류 이미지 분류 작업에서 대부분의 분류 모형은 의류 이미지 자체 또는 전문모델 착용 의류와 같이 통제된 상황에서 촬영되는 온라인 제품 이미지를 사용하여 학습을 수행한다. 하지만 본 연구에서는 통제되지 않은 상황에서 촬영되고 사람들의 움직임과 다양한 포즈가 포함된 스트릿 패션 이미지 또는 런웨이 이미지를 분류하려는 상황을 고려하여 분류 모형을 훈련시키는 효과적인 방법을 제안한다. 이동성을 포착하는 런웨이 의류 이미지로 모형을 학습시킴으로써 분류 모형의 다양한 쿼리 이미지에 대한 적응력을 높일 수 있다. 모형 학습 시 먼저 ImageNet 데이터셋을 사용하여 pre-training 과정을 거치고 본 연구를 위해 수집된 32 개 주요 패션 브랜드의 2426개 런웨이 이미지로 구성된 데이터셋을 사용하여 fine-tuning을 수행한다. 학습 과정의 일반화를 고려해 10번의 실험을 수행하고 제안된 모형은 최종 테스트에서 67.2 %의 정확도를 기록했다. 본 연구 모형은 쿼리 이미지가 런웨이 이미지, 제품 이미지 또는 스트릿 패션 이미지가 될 수 있는 다양한 분류 환경에 적용될 수 있다. 구체적으로는 패션 위크에서 모바일 어플리케이션 서비스를 통해 브랜드 검색을 용이하게 하는 서비스를 제공하거나, 패션 잡지사의 편집 작업에 사용되어 브랜드나 스타일을 분류하고 라벨을 붙일 수 있으며, 온라인 쇼핑몰에서 아이템 정보를 제공하거나 유사한 아이템을 추천하는 등의 다양한 목적에 적용될 수 있다.

• Asia pacific journal of information systems
• /
• 제17권4호
• /
• pp.187-206
• /
• 2007

The development of computer and information technology has been combined with the information superhighway internet infrastructure, so information widely spreads not only in special fields but also in the daily lives of people. Information ubiquity influences the traditional way of transaction, and leads a new E-commerce which distinguishes from the existing E-commerce. Not only goods as physical but also service as non-physical come into E-commerce. As the scale of E-Commerce is being enlarged as well. It keeps people from finding information they want. Recommender systems are now becoming the main tools for E-Commerce to mitigate the information overload. Recommender systems can be defined as systems for suggesting some Items(goods or service) considering customers' interests or tastes. They are being used by E-commerce web sites to suggest products to their customers who want to find something for them and to provide them with information to help them decide which to purchase. There are several approaches of recommending goods to customer in recommender system but in this study, the main subject is focused on collaborative filtering technique. This study presents a possibility of pre-evaluation for the prediction performance of customer's preference in collaborative filtering before the process of customer's preference prediction. Pre-evaluation for the prediction performance of each customer having low performance is classified by using the statistical features of ratings rated by each customer is conducted before the prediction process. In this study, MovieLens 100K dataset is used to analyze the accuracy of classification. The classification criteria are set by using the training sets divided 80% from the 100K dataset. In the process of classification, the customers are divided into two groups, classified group and non classified group. To compare the prediction performance of classified group and non classified group, the prediction process runs the 20% test set through the Neighborhood Based Collaborative Filtering Algorithm and Correspondence Mean Algorithm. The prediction errors from those prediction algorithm are allocated to each customer and compared with each user's error. Research hypothesis : Two research hypotheses are formulated in this study to test the accuracy of the classification criterion as follows. Hypothesis 1: The estimation accuracy of groups classified according to the standard deviation of each user's ratings has significant difference. To test the Hypothesis 1, the standard deviation is calculated for each user in training set which is divided 80% from MovieLens 100K dataset. Four groups are classified according to the quartile of the each user's standard deviations. It is compared to test the estimation errors of each group which results from test set are significantly different. Hypothesis 2: The estimation accuracy of groups that are classified according to the distribution of each user's ratings have significant differences. To test the Hypothesis 2, the distributions of each user's ratings are compared with the distribution of ratings of all customers in training set which is divided 80% from MovieLens 100K dataset. It assumes that the customers whose ratings' distribution are different from that of all customers would have low performance, so six types of different distributions are set to be compared. The test groups are classified into fit group or non-fit group according to the each type of different distribution assumed. The degrees in accordance with each type of distribution and each customer's distributions are tested by the test of ${\chi}^2$ goodness-of-fit and classified two groups for testing the difference of the mean of errors. Also, the degree of goodness-of-fit with the distribution of each user's ratings and the average distribution of the ratings in the training set are closely related to the prediction errors from those prediction algorithms. Through this study, the customers who have lower performance of prediction than the rest in the system are classified by those two criteria, which are set by statistical features of customers ratings in the training set, before the prediction process.

• 한국정보과학회 언어공학연구회:학술대회논문집(한글 및 한국어 정보처리)
• /
• 한국정보과학회언어공학연구회 2023년도 제35회 한글 및 한국어 정보처리 학술대회
• /
• pp.603-608
• /
• 2023

본 연구는 한국어 Vision-Language Pre-training 모델 학습을 위한 대규모 시각-언어 멀티모달 데이터셋 구축에 대한 필요성을 연구한다. 현재, 한국어 시각-언어 멀티모달 데이터셋은 부족하며, 양질의 데이터 획득이 어려운 상황이다. 따라서, 본 연구에서는 기계 번역을 활용하여 외국어(영문) 시각-언어 데이터를 한국어로 번역하고 이를 기반으로 생성형 AI를 활용한 데이터셋 구축 방법론을 제안한다. 우리는 다양한 캡션 생성 방법 중, ChatGPT를 활용하여 자연스럽고 고품질의 한국어 캡션을 자동으로 생성하기 위한 새로운 방법을 제안한다. 이를 통해 기존의 기계 번역 방법보다 더 나은 캡션 품질을 보장할 수 있으며, 여러가지 번역 결과를 앙상블하여 멀티모달 데이터셋을 효과적으로 구축하는데 활용한다. 뿐만 아니라, 본 연구에서는 의미론적 유사도 기반 평가 방식인 캡션 투영 일치도(Caption Projection Consistency) 소개하고, 다양한 번역 시스템 간의 영-한 캡션 투영 성능을 비교하며 이를 평가하는 기준을 제시한다. 최종적으로, 본 연구는 ChatGPT를 이용한 한국어 멀티모달 이미지-텍스트 멀티모달 데이터셋 구축을 위한 새로운 방법론을 제시하며, 대표적인 기계 번역기들보다 우수한 영한 캡션 투영 성능을 증명한다. 이를 통해, 우리의 연구는 부족한 High-Quality 한국어 데이터 셋을 자동으로 대량 구축할 수 있는 방향을 보여주며, 이 방법을 통해 딥러닝 기반 한국어 Vision-Language Pre-training 모델의 성능 향상에 기여할 것으로 기대한다.

• 한국정보통신학회논문지
• /
• 제25권11호
• /
• pp.1505-1511
• /
• 2021

자연어 처리 분야 내 다양한 작업들에서 높은 성능을 보인 사전 학습된 모델은 대량의 말뭉치를 이용하여 문장들의 언어학적 패턴을 스스로 학습함으로써 입력 문장 내 각 토큰들을 적절한 특징 벡터로 표현할 수 있다는 장점을 갖고 있다. 이러한 사전 학습된 모델의 학습에 필요한 말뭉치를 구축하는 방법 중 웹 크롤러를 이용하여 수집한 경우 웹사이트에 존재하는 문장은 다양한 패턴을 갖고 있기 때문에 문장의 일부 또는 전체에 불필요한 단어가 포함되어 있을 수 있다. 본 논문에서는 웹으로부터 수집한 말뭉치에 대해 신경망 모델을 이용하여 불필요한 단어가 포함된 문장을 필터링하기 위한 데이터 셋 구축 방법에 대해 제안한다. 그 결과, 총 2,330개의 문장을 포함한 데이터 셋을 구축하였다. 또한 신경망 모델을 이용하여 구축한 데이터 셋을 학습시켜 성능을 평가하였으며, BERT 모델이 평가 데이터에 대해 93.75%의 정확도로 가장 높은 성능을 보였다.

• Journal of Multimedia Information System
• /
• 제6권4호
• /
• pp.209-216
• /
• 2019

Over the past decade, researchers were able to solve complex medical problems as well as acquire deeper understanding of entire issue due to the availability of machine learning techniques, particularly predictive algorithms and automatic recognition of patterns in medical imaging. In this study, a technique called transfer learning has been utilized to classify Magnetic Resonance (MR) images by a pre-trained Convolutional Neural Network (CNN). Rather than training an entire model from scratch, transfer learning approach uses the CNN model by fine-tuning them, to classify MR images into Alzheimer's disease (AD), mild cognitive impairment (MCI) and normal control (NC). The performance of this method has been evaluated over Alzheimer's Disease Neuroimaging (ADNI) dataset by changing the learning rate of the model. Moreover, in this study, in order to demonstrate the transfer learning approach we utilize different pre-trained deep learning models such as GoogLeNet, VGG-16, AlexNet and ResNet-18, and compare their efficiency to classify AD. The overall classification accuracy resulted by GoogLeNet for training and testing was 99.84% and 98.25% respectively, which was exceptionally more than other models training and testing accuracies.

• Journal of Multimedia Information System
• /
• 제6권2호
• /
• pp.81-86
• /
• 2019

Deep learning is emerging as one of the best tool in processing data related to medical imaging. In our research work, we have proposed a deep learning based framework CNN (Convolutional Neural Network) for the classification of dysplastic tissue images. The CNN has classified the given images into 4 different classes namely normal tissue, mild dysplastic tissue, moderate dysplastic tissue and severe dysplastic tissue. The dataset under taken for the study consists of 672 tissue images of epithelial squamous layer of oral cavity captured out of the biopsy samples of 52 patients. After applying the data pre-processing and augmentation on the given dataset, 2688 images were created. Further, these 2688 images were classified into 4 categories with the help of expert Oral Pathologist. The classified data was supplied to the convolutional neural network for training and testing of the proposed framework. It has been observed that training data shows 91.65% accuracy whereas the testing data achieves 89.3% accuracy. The results produced by our proposed framework are also tested and validated by comparing the manual results produced by the medical experts working in this area.

• Soft Computing and Machine Intelligence
• /
• 제1권1호
• /
• pp.11-23
• /
• 2021

House price prediction is a significant financial decision for individuals working in the housing market as well as for potential buyers. From investment to buying a house for residence, a person investing in the housing market is interested in the potential gain. This paper presents machine learning algorithms to develop intelligent regressions models for House price prediction. The proposed research methodology consists of four stages, namely Data Collection, Pre Processing the data collected and transforming it to the best format, developing intelligent models using machine learning algorithms, training, testing, and validating the model on house prices of the housing market in the Capital, Islamabad. The data used for model validation and testing is the asking price from online property stores, which provide a reasonable estimate of the city housing market. The prediction model can significantly assist in the prediction of future housing prices in Pakistan. The regression results are encouraging and give promising directions for future prediction work on the collected dataset.

• The Korean Journal of Physiology and Pharmacology
• /
• 제24권1호
• /
• pp.89-99
• /
• 2020

Although microscopic analysis of tissue slides has been the basis for disease diagnosis for decades, intra- and inter-observer variabilities remain issues to be resolved. The recent introduction of digital scanners has allowed for using deep learning in the analysis of tissue images because many whole slide images (WSIs) are accessible to researchers. In the present study, we investigated the possibility of a deep learning-based, fully automated, computer-aided diagnosis system with WSIs from a stomach adenocarcinoma dataset. Three different convolutional neural network architectures were tested to determine the better architecture for tissue classifier. Each network was trained to classify small tissue patches into normal or tumor. Based on the patch-level classification, tumor probability heatmaps can be overlaid on tissue images. We observed three different tissue patterns, including clear normal, clear tumor and ambiguous cases. We suggest that longer inspection time can be assigned to ambiguous cases compared to clear normal cases, increasing the accuracy and efficiency of histopathologic diagnosis by pre-evaluating the status of the WSIs. When the classifier was tested with completely different WSI dataset, the performance was not optimal because of the different tissue preparation quality. By including a small amount of data from the new dataset for training, the performance for the new dataset was much enhanced. These results indicated that WSI dataset should include tissues prepared from many different preparation conditions to construct a generalized tissue classifier. Thus, multi-national/multi-center dataset should be built for the application of deep learning in the real world medical practice.