• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.2
• /
• pp.751-770
• /
• 2019

Action recognition is an essential task in computer vision due to the variety of prospective applications, such as security surveillance, machine learning, and human-computer interaction. The availability of more video data than ever before and the lofty performance of deep convolutional neural networks also make it essential for action recognition in video. Unfortunately, limited crafted video features and the scarcity of benchmark datasets make it challenging to address the multi-person action recognition task in video data. In this work, we propose a deep convolutional neural network-based Effective Hybrid Learning (EHL) framework for two-person interaction classification in video data. Our approach exploits a pre-trained network model (the VGG16 from the University of Oxford Visual Geometry Group) and extends the Faster R-CNN (region-based convolutional neural network a state-of-the-art detector for image classification). We broaden a semi-supervised learning method combined with an active learning method to improve overall performance. Numerous types of two-person interactions exist in the real world, which makes this a challenging task. In our experiment, we consider a limited number of actions, such as hugging, fighting, linking arms, talking, and kidnapping in two environment such simple and complex. We show that our trained model with an active semi-supervised learning architecture gradually improves the performance. In a simple environment using an Intelligent Technology Laboratory (ITLab) dataset from Inha University, performance increased to 95.6% accuracy, and in a complex environment, performance reached 81% accuracy. Our method reduces data-labeling time, compared to supervised learning methods, for the ITLab dataset. We also conduct extensive experiment on Human Action Recognition benchmarks such as UT-Interaction dataset, HMDB51 dataset and obtain better performance than state-of-the-art approaches.

• Journal of Intelligence and Information Systems
• /
• v.22 no.4
• /
• pp.193-215
• /
• 2016

In recent years, the proliferation of diverse social networking services has led users to use many mediums simultaneously depending on their individual purpose and taste. Besides, while collecting information about particular themes, they usually employ various mediums such as social networking services, Internet news, and blogs. However, in terms of management, each document circulated through diverse mediums is placed in different categories on the basis of each source's policy and standards, hindering any attempt to conduct research on a specific category across different kinds of sources. For example, documents containing content on "Application for a foreign travel" can be classified into "Information Technology," "Travel," or "Life and Culture" according to the peculiar standard of each source. Likewise, with different viewpoints of definition and levels of specification for each source, similar categories can be named and structured differently in accordance with each source. To overcome these limitations, this study proposes a plan for conducting category mapping between different sources with various mediums while maintaining the existing category system of the medium as it is. Specifically, by re-classifying individual documents from the viewpoint of diverse sources and storing the result of such a classification as extra attributes, this study proposes a logical layer by which users can search for a specific document from multiple heterogeneous sources with different category names as if they belong to the same source. Besides, by collecting 6,000 articles of news from two Internet news portals, experiments were conducted to compare accuracy among sources, supervised learning and semi-supervised learning, and homogeneous and heterogeneous learning data. It is particularly interesting that in some categories, classifying accuracy of semi-supervised learning using heterogeneous learning data proved to be higher than that of supervised learning and semi-supervised learning, which used homogeneous learning data. This study has the following significances. First, it proposes a logical plan for establishing a system to integrate and manage all the heterogeneous mediums in different classifying systems while maintaining the existing physical classifying system as it is. This study's results particularly exhibit very different classifying accuracies in accordance with the heterogeneity of learning data; this is expected to spur further studies for enhancing the performance of the proposed methodology through the analysis of characteristics by category. In addition, with an increasing demand for search, collection, and analysis of documents from diverse mediums, the scope of the Internet search is not restricted to one medium. However, since each medium has a different categorical structure and name, it is actually very difficult to search for a specific category insofar as encompassing heterogeneous mediums. The proposed methodology is also significant for presenting a plan that enquires into all the documents regarding the standards of the relevant sites' categorical classification when the users select the desired site, while maintaining the existing site's characteristics and structure as it is. This study's proposed methodology needs to be further complemented in the following aspects. First, though only an indirect comparison and evaluation was made on the performance of this proposed methodology, future studies would need to conduct more direct tests on its accuracy. That is, after re-classifying documents of the object source on the basis of the categorical system of the existing source, the extent to which the classification was accurate needs to be verified through evaluation by actual users. In addition, the accuracy in classification needs to be increased by making the methodology more sophisticated. Furthermore, an understanding is required that the characteristics of some categories that showed a rather higher classifying accuracy of heterogeneous semi-supervised learning than that of supervised learning might assist in obtaining heterogeneous documents from diverse mediums and seeking plans that enhance the accuracy of document classification through its usage.

• Korean Management Science Review
• /
• v.28 no.1
• /
• pp.53-60
• /
• 2011

Time series analysis methods have been traditionally used in stock price prediction. However, most of the existing methods represent some methodological limitations in reflecting influence from external factors that affect the fluctuation of stock prices, such as oil prices, exchange rates, money interest rates, and the stock price indexes of other countries. To overcome the limitations, we propose a network based method incorporating the relations between the individual company stock prices and the external factors by using a graph-based semi-supervised learning algorithm. For verifying the significance of the proposed method, it was applied to the prediction problems of company stock prices listed in the KOSPI from January 2007 to August 2008.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.618-620
• /
• 2022

게임은 수많은 NPC 와 규칙에 의해 작동되는 가상 공간을 의미한다. 이런 가상 공간에서는 규칙을 엄격히 지키면서 수행되는 AI 를 필수로 요구하게 된다. 하지만 강화 학습 기반의 AI 는 복잡한 게임의 규칙을 온전히 지키지 못하고 예상 밖의 행동을 돌출하면서 이를 해결하기 위한 많은 연구도 수행되고 있다. 본 논문에서는 규칙 기반으로 획득한 오목판의 확률 맵과 학습을 통해 획득한 확률맵 데이터를 병합하여 가장 높은 Value 를 가지는 위치를 다음 수로 반환하는 방법을 사용하였다. 향후 연구에서는 ANN(Approximate Nearest Neighbor)알고리즘을 적극 활용하여, 커널의 State 와 보드의 State 비교를 확률적으로 개선할 예정이다. 본 논문에서 제안된 프레임 워크는 게임 AI 연구에 기여할 수 있길 바란다.

• Korean Journal of Remote Sensing
• /
• v.34 no.6_2
• /
• pp.1239-1249
• /
• 2018

Optical remote sensing sensors provide visually more familiar images than radar images. However, it is difficult to discriminate sea ice types in optical images using spectral information based machine learning algorithms. This study addresses two topics. First, we propose a semantic segmentation which is a part of the state-of-the-art deep learning algorithms to identify ice types by learning hierarchical and spatial features of sea ice. Second, we propose a new approach by combining of semi-supervised and active learning to obtain accurate and meaningful labels from unlabeled or unseen images to improve the performance of supervised classification for multiple images. Therefore, we successfully added new labels from unlabeled data to automatically update the semantic segmentation model. This should be noted that an operational system to generate ice type products from optical remote sensing data may be possible in the near future.

• Journal of IKEEE
• /
• v.22 no.3
• /
• pp.816-821
• /
• 2018

Recently, machine learning algorithms based on artificial neural networks started to be used widely as classifiers in the field of EEG research for emotion analysis and disease diagnosis. When a machine learning model is used to classify EEG data, if training data is composed of only data having similar characteristics, classification performance may be deteriorated when applied to data of another group. In this paper, we propose a method to construct training data set by selecting several groups of data using semi-supervised learning algorithm to improve these problems. We then compared the performance of the two models by training the model with a training data set consisting of data with similar characteristics to the training data set constructed using the proposed method.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.3
• /
• pp.129-138
• /
• 2022

Collecting and labeling sufficient training data, which is essential to deep learning-based visual inspection, is difficult for manufacturers to perform because it is very expensive. This paper presents a steel plate surface defect detection system with industrial-grade detection performance by training a small amount of steel plate surface images consisting of labeled and non-labeled data. To overcome the problem of lack of training data, we propose two data augmentation techniques: program-based augmentation, which generates defect images in a geometric way, and generative model-based augmentation, which learns the distribution of labeled data. We also propose a 4-step semi-supervised learning using pseudo labels and consistency training with fixed-size augmentation in order to utilize unlabeled data for training. The proposed technique obtained about 99% defect detection performance for four defect types by using 100 real images including labeled and unlabeled data.

• Journal of Computing Science and Engineering
• /
• v.7 no.2
• /
• pp.112-121
• /
• 2013

The last years have seen a rise of interest in using electroencephalography-based brain computer interfacing methodology for investigating non-medical questions, beyond the purpose of communication and control. One of these novel applications is to examine how signal quality is being processed neurally, which is of particular interest for industry, besides providing neuroscientific insights. As for most behavioral experiments in the neurosciences, the assessment of a given stimulus by a subject is required. Based on an EEG study on speech quality of phonemes, we will first discuss the information contained in the neural correlate of this judgement. Typically, this is done by analyzing the data along behavioral responses/labels. However, participants in such complex experiments often guess at the threshold of perception. This leads to labels that are only partly correct, and oftentimes random, which is a problematic scenario for using supervised learning. Therefore, we propose a novel supervised-unsupervised learning scheme, which aims to differentiate true labels from random ones in a data-driven way. We show that this approach provides a more crisp view of the brain states that experimenters are looking for, besides discovering additional brain states to which the classical analysis is blind.

• Annual Conference on Human and Language Technology
• /
• 2013.10a
• /
• pp.45-49
• /
• 2013

기존 연구에서는 질의 응답 시스템에서 정답 유형을 분류하기 위해 패턴 매칭 방식이나 교사 학습(Supervised Learning)을 이용했다. 패턴 매칭 방식은 질의 분석을 통해 수동으로 패턴을 구축해야 한다. 교사 학습에서는 훈련 데이터 전체에 정답 유형이 태깅(Tagging)되어야 하며, 이를 위해서는 사용자의 질의에 정답 유형을 수동으로 태깅하는 작업이 많이 필요하다. 웹을 통해 정답 유형이 태깅되지 않은 대용량의 사용자 질의 말뭉치를 구할 수 있지만, 이 데이터에는 정답 유형이 태깅되어 있지 않다. 따라서, 대용량의 사용자 질의에 비례하여, 정답 유형을 수동으로 태깅하는 작업량이 증가한다. 앞서 언급한 두 가지 방법론에서, 정답 유형 분류를 위해 수작업이 많이 필요하다는 문제점을 해결하고자 본 논문에서는 일부 태깅된 훈련 데이터를 필요로 하는 반교사 학습(Semi-supervised Learning)에 기반한 정답 유형 분류를 제안한다. 이는 정답 유형 분류 작업에 필요한 노동력을 최소화함으로 대용량의 데이터를 통한 효율적 질의 응답 시스템 구축을 가능하게 한다.

• Science of Emotion and Sensibility
• /
• v.18 no.1
• /
• pp.103-110
• /
• 2015

Sentiment lexicon is an essential element for expressing sentiment on a text or recognizing sentiment from a text. We propose a graph-based semi-supervised learning method to construct a sentiment dictionary as sentiment lexicon set. In particular, we focus on the construction of domain-specific sentiment dictionary. The proposed method makes up a graph according to lexicons and proximity among lexicons, and sentiments of some lexicons which already know their sentiment values are propagated throughout all of the lexicons on the graph. There are two typical types of the sentiment lexicon, sentiment words and sentiment phrase, and we construct a sentiment dictionary by creating each graph of them and infer sentiment of all sentiment lexicons. In order to verify our proposed method, we constructed a sentiment dictionary specific to the movie domain, and conducted sentiment classification experiments with it. As a result, it have been shown that the classification performance using the sentiment dictionary is better than the other using typical general-purpose sentiment dictionary.