• Journal of Power Electronics
• /
• v.16 no.3
• /
• pp.1097-1109
• /
• 2016

Fault detection and isolation are related to system monitoring, identifying when a fault has occurred, and determining the type of fault and its location. Fault detection is utilized to determine whether a problem has occurred within a certain channel or area of operation. Fault detection and diagnosis have become increasingly important for many technical processes in the development of safe and efficient advanced systems for supervision. This paper presents an integrated technique for fault diagnosis and classification for open- and short-circuit faults in three-phase inverter circuits. Discrete wavelet transform and principal component analysis are utilized to detect the discontinuity in currents caused by a fault. The features of fault diagnosis are then extracted. A fault dictionary is used to acquire details about transistor faults and the corresponding fault identification. Fault classification is performed with a fuzzy logic system and relevance vector machine (RVM). The proposed model is incorporated with a set of optimization techniques, namely, evolutionary particle swarm optimization (EPSO) and cuckoo search optimization (CSO), to improve fault detection. The combination of optimization techniques with classification techniques is analyzed. Experimental results confirm that the combination of CSO with RVM yields better results than the combinations of CSO with fuzzy logic system, EPSO with RVM, and EPSO with fuzzy logic system.

• ETRI Journal
• /
• v.42 no.2
• /
• pp.230-238
• /
• 2020

Human-object interaction (HOI) detection is a popular computer vision task that detects interactions between humans and objects. This task can be useful in many applications that require a deeper understanding of semantic scenes. Current HOI detection networks typically consist of a feature extractor followed by detection layers comprising small filters (eg, 1 × 1 or 3 × 3). Although small filters can capture local spatial features with a few parameters, they fail to capture larger context information relevant for recognizing interactions between humans and distant objects owing to their small receptive regions. Hence, we herein propose a three-stream HOI detection network that employs a context convolution module (CCM) in each stream branch. The CCM can capture larger contexts from input feature maps by adopting combinations of large separable convolution layers and residual-based convolution layers without increasing the number of parameters by using fewer large separable filters. We evaluate our HOI detection method using two benchmark datasets, V-COCO and HICO-DET, and demonstrate its state-of-the-art performance.

• International Journal of Advanced Culture Technology
• /
• v.7 no.4
• /
• pp.321-326
• /
• 2019

As AI technology is recently introduced into various fields, it is being applied to the fashion field. This paper proposes a system for recommending cody clothes suitable for a user's selected clothes. The proposed system consists of user app, cody recommendation module, and server interworking of each module and managing database data. Cody recommendation system classifies clothing images into 80 categories composed of feature combinations, selects multiple representative reference images for each category, and selects 3 full body cordy images for each representative reference image. Cody images of the representative reference image were determined by analyzing the user's preference using Google survey app. The proposed algorithm classifies categories the clothing image selected by the user into a category, recognizes the most similar image among the classification category reference images, and transmits the linked cody images to the user's app. The proposed system uses the ResNet-50 model to categorize the input image and measures similarity using ORB and HOG features to select a reference image in the category. We test the proposed algorithm in the Android app, and the result shows that the recommended system runs well.

• Journal of architectural history
• /
• v.24 no.5
• /
• pp.7-20
• /
• 2015

Drago $Gali{\acute{c}}$ (1907-1992) has been acknowledged as one of most important modern Croatia architects of the 20th century and noted for his controversial apartment buildings at 35-35a and 43-43a blocks on Vukovar Street in Zagreb, Croatia. Although the two housings were highly regarded as the best examples of the post-war housing design in Croatia, a plagiarism controversy arose due to its similar exterior looks to Le Corbusier's $Unit{\acute{e}}$ d'Habitation in Marseille in 1952. This research intends to comparatively analyze architectural features implemented on the works of apartment of Drago $Gali{\acute{c}}$ and Le Corbusier's $Unit{\acute{e}}$ d'Habitation. The analysis focuses on architectural characteristics categorized in three parts: unit plan, community space, and unit combinations. The site survey was carried out to yield more useful information for the analysis. During this process, written and photographic documentations are collected for the further interpretation. In addition, scale drawings are reconstructed for the in-depth analysis of the project.

• KIISE Transactions on Computing Practices
• /
• v.21 no.8
• /
• pp.567-571
• /
• 2015

Transition-based dependency parsing requires much time and efforts to design and select features from a very large number of possible combinations. Recent studies have successfully applied Multi-Layer Perceptrons (MLP) to find solutions to this problem and to reduce the data sparseness. However, most of these methods have adopted greedy search and can only consider a limited amount of information from the context window. In this study, we use a Recurrent Neural Network to handle long dependencies between sub dependency trees of current state and current transition action. The results indicate that our method provided a higher accuracy (UAS) than an MLP based model.

• Journal of Information Processing Systems
• /
• v.17 no.2
• /
• pp.369-384
• /
• 2021

In a personalized product recommendation system, when the amount of log data is large or sparse, the accuracy of model recommendation will be greatly affected. To solve this problem, a personalized product recommendation method using deep factorization machine (DeepFM) to analyze user behavior is proposed. Firstly, the K-means clustering algorithm is used to cluster the original log data from the perspective of similarity to reduce the data dimension. Then, through the DeepFM parameter sharing strategy, the relationship between low- and high-order feature combinations is learned from log data, and the click rate prediction model is constructed. Finally, based on the predicted click-through rate, products are recommended to users in sequence and fed back. The area under the curve (AUC) and Logloss of the proposed method are 0.8834 and 0.0253, respectively, on the Criteo dataset, and 0.7836 and 0.0348 on the KDD2012 Cup dataset, respectively. Compared with other newer recommendation methods, the proposed method can achieve better recommendation effect.

• Nuclear Engineering and Technology
• /
• v.54 no.11
• /
• pp.4359-4372
• /
• 2022

Turbulent mixing in buoyant flows is an essential mechanism involved in many scenarios related to nuclear safety in nuclear power plants. Comprehensive understanding and accurate predictions of turbulent buoyant flows in the reactor are of crucial importance, due to the function of mitigating the potential detrimental consequences during postulated accidents. The present study uses URANS methodology to investigate the buoyancy-influenced flows in the reactor pressure vessel under the main steam line break accident scenarios. With a particular focus on the influence of turbulent heat flux closure models, various combinations of two turbulence models and three turbulent heat flux models are utilized for the numerical simulations of three ROCOM tests which have different characteristic features in terms of the flow rate and fluid density difference between loops. The simulation results are compared with experimental measurements of the so-called mixing scalar in the downcomer and at the core inlet. The study shows that the anisotropic turbulent heat flux models are able to improve the accuracy of the predictions under conditions of strong buoyancy whilst in the weak buoyancy case, a major role is played by the selected turbulence models with essentially a negligible influence of the turbulent heat flux closure models.

• The Journal of Information Systems
• /
• v.32 no.2
• /
• pp.87-108
• /
• 2023

Purpose The main purpose of this study is to improve fake news detection performance by using video information to overcome the limitations of extant text- and image-oriented studies that do not reflect the latest news consumption trend. Design/methodology/approach This study collected video clips and related information including news scripts, speakers' facial expression, and video metadata from YouTube to develop fake news detection model. Based on the collected data, seven combinations of related information (i.e. scripts, video metadata, facial expression, scripts and video metadata, scripts and facial expression, and scripts, video metadata, and facial expression) were used as an input for taining and evaluation. The input data was analyzed using six models such as support vector machine and deep neural network. The area under the curve(AUC) was used to evaluate the performance of classification model. Findings The results showed that the ACU and accuracy values of three features combination (scripts, video metadata, and facial expression) were the highest in logistic regression, naïve bayes, and deep neural network models. This result implied that the fake news detection could be improved by using video information(video metadata and facial expression). Sample size of this study was relatively small. The generalizablity of the results would be enhanced with a larger sample size.

• Journal of the Korean Academy of Child and Adolescent Psychiatry
• /
• v.28 no.2
• /
• pp.132-140
• /
• 2017

Objectives: Very early-onset schizophrenia (VEOS) is a type of psychosis having a low frequency, insidious onset, and devastating clinical outcome. In this study, the demographic features, information on medication, clinical outcomes, and intellectual capability of patients diagnosed with VEOS in a hospital were analyzed to provide therapeutic strategies for this type of schizophrenia. Methods: Using the electronic medical records of the National Center for Mental Health, 69 patients with VEOS were identified based on the DSM-5 criteria of schizophrenia. The data were summarized and analyzed according to the demographic characteristics, medications used, intellectual strength measured by the full intelligence quotient (FIQ) score, and current clinical status measured by the Clinical Global Impression-Severity (CGI-S) and various combinations of these parameters. Results: The screened study group contained similar numbers of males and females. The younger the onset of psychosis, the lower the frequency. The study population included a significantly higher proportion of births in the winter season than that of the general population. The 3 most frequently used antipsychotic medications were risperidone and its derivatives, clozapine and olanzapine. Valproic acid and divalproex sodium were the most commonly added drugs for outcome augmentation. 53.5% of the study population had received benzodiazepines and/or hypnotics. The average FIQ of the study population was 69.4, which is quite low compared to previous Korean studies with similar populations. There was a weak negative correlation between FIQ and CGI-S, but it was not statistically significant. The average CGI-S score was 4.2, which meant that the patients were moderately ill. Conclusion: This study demonstrated that patients with VEOS showed more frequent intellectual deficits at baseline and poorer outcomes than the control group. Risperidone, clozapine, valproic acid and their combinations were the most preferred medications for the treatment of psychosis. Benzodiazepines were quite commonly added for various reasons.

• Journal of KIISE:Software and Applications
• /
• v.35 no.6
• /
• pp.347-356
• /
• 2008

High-throughput microarray is one of the most popular tools in molecular biology, and various computational methods have been developed for the microarray data analysis. While the computational methods easily extract significant features, it suffers from inferring modules of multiple co-regulated genes. Hypernetworhs are motivated by biological networks, which handle all elements based on their combinatorial processes. Hence, the hypernetworks can naturally analyze the biological effects of gene combinations. In this paper, we introduce a hypernetwork classifier for microRNA (miRNA) profile analysis based on microarray data. The hypernetwork classifier uses miRNA pairs as elements, and an evolutionary learning is performed to model the microarray profiles. miTNA modules are easily extracted from the hypernetworks, and users can directly evaluate if the miRNA modules are significant. For experimental results, the hypernetwork classifier showed 91.46% accuracy for miRNA expression profiles on multiple human canters, which outperformed other machine learning methods. The hypernetwork-based analysis showed that our approach could find biologically significant miRNA modules.