• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.41 no.8
• /
• pp.958-964
• /
• 2016

In this paper, we propose an i-vector based language recognition system to identify the spoken language of the speaker, which uses multiple discriminative deep neural network (DNN) models analogous to the multi-class support vector machine (SVM) classification system. The proposed model was trained and tested using the i-vectors included in the NIST 2015 i-vector Machine Learning Challenge database, and shown to outperform the conventional language recognition methods such as cosine distance, SVM and softmax NN classifier in open-set experiments.

• Smart Structures and Systems
• /
• v.26 no.6
• /
• pp.797-807
• /
• 2020

Due to the influence of nonlinearity and time-variation, it is difficult to establish an accurate model of concrete frame structures that adopt active controllers. Fuzzy theory is a relatively appropriate method but susceptible to human subjective experience to decrease the performance. This paper proposes a novel artificial intelligence based EBA (Evolved Bat Algorithm) controller with machine learning matched membership functions in the complex nonlinear system. The proposed affine transformed membership functions are adopted and stabilization and performance criterion of the closed-loop fuzzy systems are obtained through a new parametrized linear matrix inequality which is rearranged by machine learning affine matched membership functions. The trajectory of the closed-loop dithered system and that of the closed-loop fuzzy relaxed system can be made as close as desired. This enables us to get a rigorous prediction of stability of the closed-loop dithered system by establishing that of the closed-loop fuzzy relaxed system.

• The KIPS Transactions:PartD
• /
• v.18D no.5
• /
• pp.329-338
• /
• 2011

Terminology recognition system which is a preceding research for text mining, information extraction, information retrieval, semantic web, and question-answering has been intensively studied in limited range of domains, especially in bio-medical domain. We propose a domain independent terminology recognition system based on machine learning method using dictionary, syntactic features, and Web search results, since the previous works revealed limitation on applying their approaches to general domain because their resources were domain specific. We achieved F-score 80.8 and 6.5% improvement after comparing the proposed approach with the related approach, C-value, which has been widely used and is based on local domain frequencies. In the second experiment with various combinations of unithood features, the method combined with NGD(Normalized Google Distance) showed the best performance of 81.8 on F-score. We applied three machine learning methods such as Logistic regression, C4.5, and SVMs, and got the best score from the decision tree method, C4.5.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.29 no.4
• /
• pp.775-784
• /
• 2019

Malware infringement attacks are continuously increasing in various environments such as mobile, IOT, windows and mac due to the emergence of new and variant malware, and signature-based countermeasures have limitations in detection of malware. In addition, analytical performance is deteriorating due to obfuscation, packing, and anti-VM technique. In this paper, we propose a system that can detect malware based on machine learning by using similarity hashing-based pattern detection technique and static analysis after file classification according to packing. This enables more efficient detection because it utilizes both pattern-based detection, which is well-known malware detection, and machine learning-based detection technology, which is advantageous for detecting new and variant malware. The results of this study were obtained by detecting accuracy of 95.79% or more for benign sample files and malware sample files provided by the AI-based malware detection track of the Information Security R&D Data Challenge 2018 competition. In the future, it is expected that it will be possible to build a system that improves detection performance by applying a feature vector and a detection method to the characteristics of a packed file.

• Genomics & Informatics
• /
• v.16 no.4
• /
• pp.31.1-31.7
• /
• 2018

The prevalence of metabolic syndrome (MS) in the nonobese population is not low. However, the identification and risk mitigation of MS are not easy in this population. We aimed to develop an MS prediction model using genetic and clinical factors of nonobese Koreans through machine learning methods. A prediction model for MS was designed for a nonobese population using clinical and genetic polymorphism information with five machine learning algorithms, including naïve Bayes classification (NB). The analysis was performed in two stages (training and test sets). Model A was designed with only clinical information (age, sex, body mass index, smoking status, alcohol consumption status, and exercise status), and for model B, genetic information (for 10 polymorphisms) was added to model A. Of the 7,502 nonobese participants, 647 (8.6%) had MS. In the test set analysis, for the maximum sensitivity criterion, NB showed the highest sensitivity: 0.38 for model A and 0.42 for model B. The specificity of NB was 0.79 for model A and 0.80 for model B. In a comparison of the performances of models A and B by NB, model B (area under the receiver operating characteristic curve [AUC] = 0.69, clinical and genetic information input) showed better performance than model A (AUC = 0.65, clinical information only input). We designed a prediction model for MS in a nonobese population using clinical and genetic information. With this model, we might convince nonobese MS individuals to undergo health checks and adopt behaviors associated with a preventive lifestyle.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.4
• /
• pp.89-96
• /
• 2021

The number of confirmed cases of coronavirus-19 is soaring around the world and has caused numerous deaths. Wearing a mask is very important to prevent infection. Incidents and accidents have occurred due to the recommendation to wear a mask in public places such as buses and subways, and it has emerged as a serious social problem. To solve this problem, this paper proposes an open source-based face mask wearing monitoring system. We used open source software, web-based artificial intelligence tool teachable machine and open source hardware Arduino. It judges whether the mask is worn, and performs commands such as guidance messages and alarms. The learning parameters of the teachable machine were learned with the optimal values of 50 learning times, 32 batch sizes, and 0.001 learning rate, resulting in an accuracy of 1 and a learning error of 0.003. We designed and implemented a mask wearing monitoring system that can perform commands such as guidance messages and alarms by determining whether to wear a mask using a web-based artificial intelligence tool teachable machine and Arduino to prove its validity.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.05a
• /
• pp.235-237
• /
• 2022

Video content is divided into long-form video content and short-form video content according to the length. Long form video content is created with a length of 15 minutes or longer, and all frames of the captured video are included without editing. Short-form video content can be edited to a shorter length from 1 minute to 15 minutes, and only some frames from the frames of the captured video. Due to the recent growth of the single-person broadcasting market, the demand for short-form video content to increase viewers is increasing. Therefore, there is a need for research on content editing technology for editing and generating short-form video content. This study studies the technology to create short-form videos of main scenes by capturing images, voices, and motions. Short-form videos of key scenes use a pre-trained highlight extraction model through machine learning. An automatic video editing system and method for automatically generating a highlight video is a core technology of short-form video content. Machine learning-based automatic video editing method and system research will contribute to competitive content activities by reducing the effort and cost and time invested by single creators for video editing

• Journal of Hydrogen and New Energy
• /
• v.31 no.5
• /
• pp.436-443
• /
• 2020

The performance prediction model of a solid oxide fuel cell stack has been developed using deep neural network technique, one of the machine learning methods. The machine learning has been received much interest in various fields, including energy system mo- deling. Using machine learning technique can save time and cost requried in developing an energy system model being compared to the conventional method, that is a combination of a mathematical modeling and an experimental validation. Results reveal that the mean average percent error, root mean square error, and coefficient of determination (R2) range 1.7515, 0.1342, 0.8597, repectively, in maximum. To improve the predictability of the model, the pre-processing is effective and interpolative machine learning and application is more accurate than the extrapolative cases.

• Smart Media Journal
• /
• v.8 no.1
• /
• pp.74-81
• /
• 2019

Recently, machine learning technology has had a significant impact on society, particularly in the medical, manufacturing, marketing, finance, broadcasting, and agricultural aspects of human lives. In this paper, we study how to apply machine learning techniques to foods, which have the greatest influence on the human survival. In the field of Smart Farm, which integrates the Internet of Things (IoT) technology into agriculture, we focus on optimizing the crop growth environment by monitoring the growth environment in real time. KT Smart Farm Solution 2.0 has adopted machine learning to optimize temperature and humidity in the greenhouse. Most existing smart farm businesses mainly focus on controlling the growth environment and improving productivity. On the other hand, in this study, we are studying how to apply machine learning with respect to harvest time so that we will be able to harvest fruits of the highest quality and ship them at an excellent cost. In order to apply machine learning techniques to the field of smart farms, it is important to acquire abundant voluminous data. Therefore, to apply accurate machine learning technology, it is necessary to continuously collect large data. Therefore, the color, value, internal temperature, and moisture of greenhouse-grown fruits are collected and secured in real time using color, weight, and temperature/humidity sensors. The proposed FPSML provides an architecture that can be used repeatedly for a similar fruit crop. It allows for a more accurate harvest time as massive data is accumulated continuously.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.6
• /
• pp.1089-1098
• /
• 2020

In this study, we propose a method to detect red tide Cochlodinium Polykrikoide using by machine learning and geostationary marine satellite images. To learn the machine learning model, GOCI Level 2 data were used, and the red tide location data of the National Fisheries Research and Development Institute was used. The machine learning model used logistic regression model, decision tree model, and random forest model. As a result of the performance evaluation, compared to the traditional GOCI image-based red tide detection algorithm without machine learning (Son et al., 2012) (75%), it was confirmed that the accuracy was improved by about 13~22%p (88~98%). In addition, as a result of comparing and analyzing the detection performance between machine learning models, the random forest model (98%) showed the highest detection accuracy.It is believed that this machine learning-based red tide detection algorithm can be used to detect red tide early in the future and track and monitor its movement and spread.