• 한국멀티미디어학회논문지
• /
• 제25권1호
• /
• pp.52-60
• /
• 2022

Few-shot learning is a sub-area of machine learning problems, which aims to classify target images that only contain a few labeled samples for training. As a representative few-shot learning method, the Prototypical network has been received much attention due to its simplicity and promising results. However, the Prototypical network uses the sample mean of samples from the same class as the prototypes of that class, which easily results in learning uncharacteristic features in the low-data scenery. In this study, we propose to use local descriptors (i.e., patches along the channel within feature maps) from the same class to explicitly obtain more representative prototypes for Prototypical Network so that significant intra-class feature information can be maintained and thus improving the classification performance on few-shot learning tasks. Experimental results on various benchmark datasets including mini-ImageNet, CUB-200-2011, and tiered-ImageNet show that the proposed method can learn more discriminative intra-class features by the local descriptors and obtain more generic prototype representations under the few-shot setting.

• 한국콘텐츠학회논문지
• /
• 제18권11호
• /
• pp.447-454
• /
• 2018

본 논문은 컨볼루션 신경망으로 이루어진 학습 모델을 통해 스테레오 영상의 깊이영상 생성 알고리즘을 제안한다. 제안하는 알고리즘은 좌, 우 시차 영상을 입력으로 받아 각 시차영상의 주요 특징을 추출하는 특징 추출부와 추출된 특징을 이용하여 시차 정보를 학습하는 깊이 학습부로 구성된다. 우선 특징 추출부는 2D CNN 계층들로 이루어진 익셉션 모듈(xception module) 및 ASPP 모듈(atrous spatial pyramid pooling) module을 통해 각각의 시차영상에 대한 특징맵을 추출한다. 그 후 각 시차에 대한 특징 맵을 시차에 따라 3차원 형태로 쌓아 3D CNN을 통해 깊이 추정 가중치를 학습하는 깊이 학습부를 거친 후 깊이 영상을 추정한다. 제안하는 알고리즘은 객체 영역에 대해 기존의 다른 학습 알고리즘들 보다 정확한 깊이를 추정하였다.

• 방송공학회논문지
• /
• 제25권7호
• /
• pp.1081-1094
• /
• 2020

With the recent development of deep learning, most computer vision-related tasks are being solved with deep learning-based network technologies such as CNN and RNN. Computer vision tasks such as object detection or object segmentation use intermediate features extracted from the same backbone such as Resnet or FPN for training and inference for object detection and segmentation. In this paper, an experiment was conducted to find out the compression efficiency and the effect of encoding on task inference performance when the features extracted in the intermediate stage of CNN are encoded. The feature map that combines the features of 256 channels into one image and the original image were encoded in HEVC to compare and analyze the inference performance for object detection and segmentation. Since the intermediate feature map encodes the five levels of feature maps (P2 to P6), the image size and resolution are increased compared to the original image. However, when the degree of compression is weakened, the use of feature maps yields similar or better inference results to the inference performance of the original image.

• 한국재료학회지
• /
• 제33권4호
• /
• pp.164-174
• /
• 2023

The bandgap characteristics of semiconductor materials are an important factor when utilizing semiconductor materials for various applications. In this study, based on data provided by AFLOW (Automatic-FLOW for Materials Discovery), the bandgap of a semiconductor material was predicted using only the material's compositional features. The compositional features were generated using the python module of 'Pymatgen' and 'Matminer'. Pearson's correlation coefficients (PCC) between the compositional features were calculated and those with a correlation coefficient value larger than 0.95 were removed in order to avoid overfitting. The bandgap prediction performance was compared using the metrics of R² score and root-mean-squared error. By predicting the bandgap with randomforest and xgboost as representatives of the ensemble algorithm, it was found that xgboost gave better results after cross-validation and hyper-parameter tuning. To investigate the effect of compositional feature selection on the bandgap prediction of the machine learning model, the prediction performance was studied according to the number of features based on feature importance methods. It was found that there were no significant changes in prediction performance beyond the appropriate feature. Furthermore, artificial neural networks were employed to compare the prediction performance by adjusting the number of features guided by the PCC values, resulting in the best R² score of 0.811. By comparing and analyzing the bandgap distribution and prediction performance according to the material group containing specific elements (F, N, Yb, Eu, Zn, B, Si, Ge, Fe Al), various information for material design was obtained.

• 대한전기학회논문지:시스템및제어부문D
• /
• 제52권2호
• /
• pp.102-106
• /
• 2003

This paper describes a new pattern classifier neural network to extract the feature from a letter. The proposed pattern classifier is based on relative distance, which is measure between an input datum and the center of cluster group. So, the proposed classifier neural network is called relative neural network(RNN). According to definitions of the distance and the learning rule, the structure of RNN is designed and the pseudo code of the algorithm is described. In feature extraction of letter, RNN, in spite of deletion of learning rate, resulted in the identical performance with those of winner-take-all(WTA), and self-organizing-map(SOM) neural network. Thus, it is shown that RNN is suitable to extract the feature of a letter.

• 한국컴퓨터정보학회논문지
• /
• 제23권11호
• /
• pp.25-30
• /
• 2018

Pattern classification system is often an important component of intelligent systems. In this paper, we present a pattern classification system consisted of the feature selection module, knowledge base construction module and decision module. We introduce a feature impact evaluation selection method based on fuzzy cluster analysis considering computational approach and generalization capability of given data characteristics. A fuzzy neural network, OFUN-NET based on unsupervised learning data mining technique produces knowledge base for representative clusters. 240 blemish pattern images are prepared and applied to the proposed system. Experimental results show the feasibility of the proposed classification system as an automating defect inspection tool.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권4호
• /
• pp.2162-2179
• /
• 2019

The performance of text classification is highly related to the feature selection methods. Usually, two tasks are performed when a feature selection method is applied to construct a feature set; 1) assign score to each feature and 2) select the top-N features. The selection of top-N features in the existing filter-based feature selection methods is biased by their discriminative power and the empirical process which is followed to determine the value of N. In order to improve the text classification performance by presenting a more illustrative feature set, we present an approach via a potent representation learning technique, namely DBN (Deep Belief Network). This algorithm learns via the semantic illustration of documents and uses feature vectors for their formulation. The nodes, iteration, and a number of hidden layers are the main parameters of DBN, which can tune to improve the classifier's performance. The results of experiments indicate the effectiveness of the proposed method to increase the classification performance and aid developers to make effective decisions in certain domains.

• 인터넷정보학회논문지
• /
• 제15권6호
• /
• pp.117-124
• /
• 2014

모바일 기기를 사용한 실시간 비디오 영상처리분야의 중요 객체탐색 및 추적의 문제에 있어서 난제는 복잡한 배경속에서 전경을 구분해 내는 일이다. 본 논문에서는 기계학습을 위한 특성벡터 선정의 문제를 위한 문맥인식 모델을 제시하여 잡음제거를 위한 기계학습기반의 구분자를 구현하였다. 수학적으로 NP-hard로 알려진 가장 가까운 이웃을 사용한 문맥인식 특성벡터 선정 알고리즘의 구현에 있어서, 본 논문은 연산횟수를 줄인 유사방법론에 대해 자세히 거론하였다. 또한, 문맥인식 성격을 가미한 특성벡터 선정을 통해 얻어진 특성 공간에서의 향상된 분리성에 대해 주성분 분석을 통해 엄밀한 분석결과를 제시하였다. 전반적인 성능 향상의 정도를 계측하기 위해 다양한 기계학습 방법론, 예를 들어, 다층신경망, 지원벡터기계, 나이브베이지안, 회귀분석 등을 사용해 비교결과를 제시하였다. 본 논문에서 제시한 방법론의 성능과 계산상 자원사용에 대한 내용을 결론으로 서술하였다.

• 한국지능정보시스템학회:학술대회논문집
• /
• 한국지능정보시스템학회 2001년도 The Pacific Aisan Confrence On Intelligent Systems 2001
• /
• pp.310-315
• /
• 2001

We are focusing on an approach which handle a general Web as a resource in order to support self-directed learning for a student. Then, we are developing a Web based learning environment "Web-Retracer"for utilizing Web as teaching materials by a user′s Annotation. Although the learner can share the Web resource that the others utilized in this environment, Web resources unsuitable for a student′s needs becomes hindrance about her/his self-directed learning. In this paper, we propose a recommending method of the resource united with a student′s needs on the basis of a student′s learning and Web browsing history. This method analyzed the feature peculiar to a resource, and extracts the resource with which the needs of the feature and a student agreed.

• 한국컴퓨터정보학회논문지
• /
• 제24권9호
• /
• pp.21-27
• /
• 2019

In this paper, we propose a deep auto-encoder-based pipe leak detection (PLD) technique from time-series acoustic data collected by microphone sensor nodes. The key idea of the proposed technique is to learn representative features of the leak-free state using leak-free time-series acoustic data and the deep auto-encoder. The proposed technique can be used to create a PLD model that detects leaks in the pipeline in an unsupervised learning manner. This means that we only use leak-free data without labeling while training the deep auto-encoder. In addition, when compared to the previous supervised learning-based PLD method that uses image features, this technique does not require complex preprocessing of time-series acoustic data owing to the unsupervised feature extraction scheme. The experimental results show that the proposed PLD method using the deep auto-encoder can provide reliable PLD accuracy even considering unsupervised learning-based feature extraction.