• Biomolecules & Therapeutics
• /
• v.27 no.3
• /
• pp.290-301
• /
• 2019

Paeonol has neuroprotective function, which could be useful for improving central nervous system disorder. The purpose of this study was to characterize the functional mechanism involved in brain transport of paeonol through blood-brain barrier (BBB). Brain transport of paeonol was characterized by internal carotid artery perfusion (ICAP), carotid artery single injection technique (brain uptake index, BUI) and intravenous (IV) injection technique in vivo. The transport mechanism of paeonol was examined using conditionally immortalized rat brain capillary endothelial cell line (TR-BBB) as an in vitro model of BBB. Brain volume of distribution (VD) of [$^3H$]paeonol in rat brain was about 6-fold higher than that of [$^{14}C$]sucrose, the vascular space marker of BBB. The uptake of [$^3H$]paeonol was concentration-dependent. Brain volume of distribution of paeonol and BUI as in vivo and inhibition of analog as in vitro studies presented significant reduction effect in the presence of unlabeled lipophilic compounds such as paeonol, imperatorin, diphenhydramine, pyrilamine, tramadol and ALC during the uptake of [$^3H$]paeonol. In addition, the uptake significantly decreased and increased at the acidic and alkaline pH in both extracellular and intracellular study, respectively. In the presence of metabolic inhibitor, the uptake reduced significantly but not affected by sodium free or membrane potential disruption. Similarly, paeonol uptake was not affected on OCTN2 or rPMAT siRNA transfection BBB cells. Interestingly. Paeonol is actively transported from the blood to brain across the BBB by a carrier mediated transporter system.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.18 no.4
• /
• pp.71-79
• /
• 2019

By using convolutional neural networks (CNNs) based on semantic segmentation, road surface damage detection has being studied. In order to generate the CNN model, it is essential to collect the input and the corresponding labeled images. Unfortunately, such collecting pairs of the dataset requires a great deal of time and costs. In this paper, we proposed a road surface damage detection technique based on semi-supervised learning using pseudo labels to mitigate such problem. The model is updated by properly mixing labeled and unlabeled datasets, and compares the performance against existing model using only labeled dataset. As a subjective result, it was confirmed that the recall was slightly degraded, but the precision was considerably improved. In addition, the $F_1-score$ was also evaluated as a high value.

• Journal of the Korea Convergence Society
• /
• v.9 no.11
• /
• pp.15-21
• /
• 2018

In this study, we try to detect anomalies on the network intrusion detection system by learning only one class. We use KDD CUP 1999 dataset, an intrusion detection dataset, which is used to evaluate classification performance. One class classification is one of unsupervised learning methods that classifies attack class by learning only normal class. When using unsupervised learning, it difficult to achieve relatively high classification efficiency because it does not use negative instances for learning. However, unsupervised learning has the advantage for classifying unlabeled data. In this study, we use one class classifiers based on support vector machines and density estimation to detect new unknown attacks. The test using the classifier based on density estimation has shown relatively better performance and has a detection rate of about 96% while maintaining a low FPR for the new attacks.

• Smart Structures and Systems
• /
• v.24 no.2
• /
• pp.243-256
• /
• 2019

Determination of the most meaningful structural modes and gaining insight into how these modes evolve are important issues for long-term structural health monitoring of the long-span bridges. To address this issue, modal parameters identified throughout the life of the bridge need to be compared and linked with each other, which is the process of mode tracking. The modal frequencies for a long-span bridge are typically closely-spaced, sensitive to the environment (e.g., temperature, wind, traffic, etc.), which makes the automated tracking of modal parameters a difficult process, often requiring human intervention. Machine learning methods are well-suited for uncovering complex underlying relationships between processes and thus have the potential to realize accurate and automated modal tracking. In this study, Gaussian mixture model (GMM), a popular unsupervised machine learning method, is employed to automatically determine and update baseline modal properties from the identified unlabeled modal parameters. On this foundation, a new mode tracking method is proposed for automated mode tracking for long-span bridges. Firstly, a numerical example for a three-degree-of-freedom system is employed to validate the feasibility of using GMM to automatically determine the baseline modal properties. Subsequently, the field monitoring data of a long-span bridge are utilized to illustrate the practical usage of GMM for automated determination of the baseline list. Finally, the continuously monitoring bridge acceleration data during strong typhoon events are employed to validate the reliability of proposed method in tracking the changing modal parameters. Results show that the proposed method can automatically track the modal parameters in disastrous scenarios and provide valuable references for condition assessment of the bridge structure.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.24 no.4
• /
• pp.382-391
• /
• 2021

Unsupervised domain adaptation often gives impressive solutions to handle domain shift of data. Most of current approaches assume that unlabeled target data to train is abundant. This assumption is not always true in practices. To tackle this issue, we propose a general solution to solve the domain gap minimization problem without any target data. Our method consists of two regularization steps. The first step is a pixel regularization by arbitrary style transfer. Recently, some methods bring style transfer algorithms to domain adaptation and domain generalization process. They use style transfer algorithms to remove texture bias in source domain data. We also use style transfer algorithms for removing texture bias, but our method depends on neither domain adaptation nor domain generalization paradigm. The second regularization step is a feature regularization by feature alignment. Adding a feature alignment loss term to the model loss, the model learns domain invariant representation more efficiently. We evaluate our regularization methods from several experiments both on small dataset and large dataset. From the experiments, we show that our model can learn domain invariant representation as much as unsupervised domain adaptation methods.

• Smart Structures and Systems
• /
• v.31 no.4
• /
• pp.365-381
• /
• 2023

The existing vision-based techniques for inspection and condition assessment of civil infrastructure are mostly manual and consequently time-consuming, expensive, subjective, and risky. As a viable alternative, researchers in the past resorted to deep learning-based autonomous damage detection algorithms for expedited post-disaster reconnaissance of structures. Although a number of automatic damage detection algorithms have been proposed, the scarcity of labeled training data remains a major concern. To address this issue, this study proposed a semi-supervised learning (SSL) framework based on consistency regularization and cross-supervision. Image data from post-earthquake reconnaissance, that contains cracks, spalling, and exposed rebars are used to evaluate the proposed solution. Experiments are carried out under different data partition protocols, and it is shown that the proposed SSL method can make use of unlabeled images to enhance the segmentation performance when limited amount of ground truth labels are provided. This study also proposes DeepLab-AASPP and modified versions of U-Net++ based on channel-wise attention mechanism to better segment the components and damage areas from images of reinforced concrete buildings. The channel-wise attention mechanism can effectively improve the performance of the network by dynamically scaling the feature maps so that the networks can focus on more informative feature maps in the concatenation layer. The proposed DeepLab-AASPP achieves the best performance on component segmentation and damage state segmentation tasks with mIoU scores of 0.9850 and 0.7032, respectively. For crack, spalling, and rebar segmentation tasks, modified U-Net++ obtains the best performance with Igou scores (excluding the background pixels) of 0.5449, 0.9375, and 0.5018, respectively. The proposed architectures win the second place in IC-SHM2021 competition in all five tasks of Project 2.

• KIPS Transactions on Computer and Communication Systems
• /
• v.12 no.5
• /
• pp.165-170
• /
• 2023

With the recent rise of blockchain technology, cryptocurrency platforms using it are increasing, and currency transactions are being actively conducted. However, crimes that abuse the characteristics of cryptocurrency are also increasing, which is a problem. In particular, phishing scams account for more than a majority of Ethereum cybercrime and are considered a major security threat. Therefore, effective phishing scams detection methods are urgently needed. However, it is difficult to provide sufficient data for supervised learning due to the problem of data imbalance caused by the lack of phishing addresses labeled in the Ethereum participating account address. To address this, this paper proposes a phishing scams detection method that uses both Trans2vec, an effective graph embedding techique considering Ethereum transaction networks, and semi-supervised learning model Tri-training to make the most of not only labeled data but also unlabeled data.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.1
• /
• pp.83-88
• /
• 2024

In this research, we propose a Semi-Supervised learning based railroad surface defect detection method. The Resnet50 model, pretrained on ImageNet, was employed for the training. Data without labels are randomly selected, and then labeled to train the ResNet50 model. The trained model is used to predict the results of the remaining unlabeled training data. The predicted values exceeding a certain threshold are selected, sorted in descending order, and added to the training data. Pseudo-labeling is performed based on the class with the highest probability during this process. An experiment was conducted to assess the overall class classification performance based on the initial number of labeled data. The results showed an accuracy of 98% at best with less than 10% labeled training data compared to the overall training data.

• Applied Microscopy
• /
• v.24 no.4
• /
• pp.61-77
• /
• 1994

The calcium-binding proteins (CaBP), parvalbumin (PV) and calbindin-D 28K (calbindin) are particularly abundant and specific in their distribution, and present in different subsets of neurons in many brain regions. Although their physiological roles in the neurons have not been elucidated, they are valuable markers of neuronal subpopulations for anatomical and developmental studies. This study is designed to characterize dorsal lateral geniculate nucleus (dLGN) neurons and axon terminals in terms of differential expression of immunoreactivity (IR) for two well-known CaBPs, PV and calbindin. The experiments were carried out on 6 adult monkeys. Monkeys were perfused under deep Nembutal anesthesia with 2% paraformaldehyde and 0.2% glutaraldehyde in 0.1M phosphate buffer. After removal, the brains were postfixed for 6-8 hr in 2% paraformaldehyde at $4^{\circ}C$ and infiltrated with 30% sucrose at $4^{\circ}C$. Thereafter, they were frozen in dry ice. Serial sections of the thalamus, at $20{\mu}m$, were made in the frontal plane with a sliding microtome. The sections were stained for PV and calbindin with indirect immunocytochemical methods. For electron microscopy, after infiltration with 30% sucrose the blocks of thalamus were serially sectioned at $50{\mu}m$ with a Vibratome in the coronal plane and stained immediately by indirect ABC methods without Triton X-100 in incubation medium. Stained sections were postfixed in 0.2% osmium tetroxide, dehydrated and flat-embedded in Spurr resin. The block was then trimmed to contain only a selected lamina or interlaminar space. The dLGN proper showed strong PV IR in fibers in all laminae and interlaminar zones. Particularly dense staining was noted in layers 1 and 2 that contain many stained fibers from optic tract. Neuronal cell body stained with PV was concentrated only in the laminae. In these laminae staining was moderate in cell bodies of all large and medium-sized neurons, and was strong in cell bodies of some small neurons together with their processes. Calbindin IR was marked in the neuronal cell body and neuropil in the S layers and interlaminar zones whereas moderate in the neuropil throughout the nucleus. Regional difference in distribution of PV and calbindin IR cell is distinct; the former is only in the laminae and the latter in both the S layer and interlaminar space. The CaBP-IR elements were confined to about $10{\mu}m$ in depth of Vibratome section. The IR product for CaBP was mainly associated with synaptic vesicle, pre- and post-synaptic membrane, and outer mitochondrial membrane and along microtubule. PV-IR was noted in various neuronal elements such as neuronal soma, dendrite, RLP, F, PSD and some myelinated or unmyelinated axons, and was not seen in the RSD and glial cells. Only a few neuronal components in dLGN was IR for calbindin and its reaction product was less dense than that of PV, and scattered throughout cytoplasm of soma of some relay neurons, and was also persent in some dendrite, myelinated axons and RLP. The RSD, F, PSD and glial elements were always non-IR for calbindin. Calbindin labelled RLP were presynaptic to unlabeled dendrite or dendritic spine and PSD. Calbindin-labeled dendrite of various sizes were always postsynaptic to unlabeled RSD, RLP or F. From this study it is suggested that dLGN cells of different functional systems and their differential projection to the visual cortex can be distinguished by differential expression of PV and calbindin.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.24 no.5
• /
• pp.482-488
• /
• 2014

This paper aims to analyze user's emotion automatically by analyzing Twitter, a representative social network service (SNS). In order to create sentiment analysis models by using machine learning techniques, sentiment labels that represent positive/negative emotions are required. However it is very expensive to obtain sentiment labels of tweets. So, in this paper, we propose a sentiment analysis model by using self-training technique in order to utilize "data without sentiment labels" as well as "data with sentiment labels". Self-training technique is that labels of "data without sentiment labels" is determined by utilizing "data with sentiment labels", and then updates models using together with "data with sentiment labels" and newly labeled data. This technique improves the sentiment analysis performance gradually. However, it has a problem that misclassifications of unlabeled data in an early stage affect the model updating through the whole learning process because labels of unlabeled data never changes once those are determined. Thus, labels of "data without sentiment labels" needs to be carefully determined. In this paper, in order to get high performance using self-training technique, we propose 3 policies for updating "data with sentiment labels" and conduct a comparative analysis. The first policy is to select data of which confidence is higher than a given threshold among newly labeled data. The second policy is to choose the same number of the positive and negative data in the newly labeled data in order to avoid the imbalanced class learning problem. The third policy is to choose newly labeled data less than a given maximum number in order to avoid the updates of large amount of data at a time for gradual model updates. Experiments are conducted using Stanford data set and the data set is classified into positive and negative. As a result, the learned model has a high performance than the learned models by using "data with sentiment labels" only and the self-training with a regular model update policy.