• Electronics and Telecommunications Trends
• /
• v.33 no.5
• /
• pp.121-134
• /
• 2018

The Von Neumann based architecture of the modern computer has dominated the computing industry for the past 50 years, sparking the digital revolution and propelling us into today's information age. Recent research focus and market trends have shown significant effort toward the advancement and application of artificial intelligence technologies. Although artificial intelligence has been studied for decades since the Turing machine was first introduced, the field has recently emerged into the spotlight thanks to remarkable milestones such as AlexNet-CNN and Alpha-Go, whose neural-network based deep learning methods have achieved a ground-breaking performance superior to existing recognition, classification, and decision algorithms. Unprecedented results in a wide variety of applications (drones, autonomous driving, robots, stock markets, computer vision, voice, and so on) have signaled the beginning of a golden age for artificial intelligence after 40 years of relative dormancy. Algorithmic research continues to progress at a breath-taking pace as evidenced by the rate of new neural networks being announced. However, traditional Von Neumann based architectures have proven to be inadequate in terms of computation power, and inherently inefficient in their processing of vastly parallel computations, which is a characteristic of deep neural networks. Consequently, global conglomerates such as Intel, Huawei, and Google, as well as large domestic corporations and fabless companies are developing dedicated semiconductor chips customized for artificial intelligence computations. The AI Processor Research Laboratory at ETRI is focusing on the research and development of super low-power AI processor chips. In this article, we present the current trends in computation platform, parallel processing, AI processor, and super-threaded AI processor research being conducted at ETRI.

• Journal of Information Technology Applications and Management
• /
• v.12 no.2
• /
• pp.15-27
• /
• 2005

Text categorization systems generally use single words (unigrams) as features. A deceptively simple algorithm for improving text categorization is investigated here, an idea previously shown not to work. It is to identify useful word pairs (bigrams) made up of adjacent unigrams. The bigrams it found, while small in numbers, can substantially raise the quality of feature sets. The algorithm was tested on two pre-classified datasets, Reuters-21578 for English and Korea-web for Korean. The results show that the algorithm was successful in extracting high quality bigrams and increased the quality of overall features. To find out the role of bigrams, we trained the Na$\"{i}$ve Bayes classifiers using both unigrams and bigrams as features. The results show that recall values were higher than those of unigrams alone. Break-even points and F1 values improved in most documents, especially when documents were classified along the large classes. In Reuters-21578 break-even points increased by 2.1%, with the highest at 18.8%, and F1 improved by 1.5%, with the highest at 3.2%. In Korea-web break-even points increased by 1.0%, with the highest at 4.5%, and F1 improved by 0.4%, with the highest at 4.2%. We can conclude that text classification using unigrams and bigrams together is more efficient than using only unigrams.

• Journal of KIISE:Software and Applications
• /
• v.34 no.11
• /
• pp.984-990
• /
• 2007

As the amount of electronic documents increases explosively, automatic text categorization methods are needed to identify those of interest. Most methods use machine learning techniques based on a word set. This paper introduces a new method, called FPTC (FP-Tree based Text Classifier). FP-Tree is a data structure used in data-mining. In this paper, a method of storing text sentence patterns in the FP-Tree structure and classifying text using the patterns is presented. In the experiments conducted, we use our algorithm with a #Mutual Information and Entropy# approach to improve performance. We also present an analysis of the algorithm via an ordinary differential categorization method.

• Journal of Information Processing Systems
• /
• v.16 no.3
• /
• pp.718-732
• /
• 2020

Conducting sentiment analysis and opinion mining are challenging tasks in natural language processing. Many of the sentiment analysis and opinion mining applications focus on product reviews, social media reviews, forums and microblogs whose reviews are topic-similar and opinion-rich. In this paper, we try to analyze the sentiments of sentences from online webcast reviews that scroll across the screen, which we call live barrages. Contrary to social media comments or product reviews, the topics in live barrages are more fragmented, and there are plenty of invalid comments that we must remove in the preprocessing phase. To extract evaluative sentiment sentences, we proposed a novel approach that clusters the barrages from the same commenter to solve the problem of scattering the information for each barrage. The method developed in this paper contains two subtasks: in the data preprocessing phase, we cluster the sentences from the same commenter and remove unavailable sentences; and we use a semi-supervised machine learning approach, the naïve Bayes algorithm, to analyze the sentiment of the barrage. According to our experimental results, this method shows that it performs well in analyzing the sentiment of online webcast barrages.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.147-147
• /
• 2018

In this article, we use an open source software library: TensorFlow, developed for the purposes of conducting very complex machine learning and deep neural network applications. However, the system is general enough to be applicable in a wide variety of other domains as well. The proposed model based on a deep neural network model, LSTM (Long Short-Term Memory) to predict the river water level at Okcheon Station of the Guem River without utilization of rainfall - forecast information. For LSTM modeling, the input data is hourly water level data for 15 years from 2002 to 2016 at 4 stations includes 3 upstream stations (Sutong, Hotan, and Songcheon) and the forecasting-target station (Okcheon). The data are subdivided into three purposes: a training data set, a testing data set and a validation data set. The model was formulated to predict Okcheon Station water level for many cases from 3 hours to 12 hours of lead time. Although the model does not require many input data such as climate, geography, land-use for rainfall-runoff simulation, the prediction is very stable and reliable up to 9 hours of lead time with the Nash - Sutcliffe efficiency (NSE) is higher than 0.90 and the root mean square error (RMSE) is lower than 12cm. The result indicated that the method is able to produce the river water level time series and be applicable to the practical flood forecasting instead of hydrologic modeling approaches.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.2
• /
• pp.383-403
• /
• 2021

With the development of mobile edge computing (MEC), some late-model application technologies, such as self-driving, augmented reality (AR) and traffic perception, emerge as the times require. Nevertheless, the high-latency and low-reliability of the traditional cloud computing solutions are difficult to meet the requirement of growing smart cars (SCs) with computing-intensive applications. Hence, this paper studies an efficient offloading decision and resource allocation scheme in collaborative vehicular edge computing networks with multiple SCs and multiple MEC servers to reduce latency. To solve this problem with effect, we propose a context-aware offloading strategy based on differential evolution algorithm (DE) by considering vehicle mobility, roadside units (RSUs) coverage, vehicle priority. On this basis, an autoregressive integrated moving average (ARIMA) model is employed to predict idle computing resources according to the base station traffic in different periods. Simulation results demonstrate that the practical performance of the context-aware vehicular task offloading (CAVTO) optimization scheme could reduce the system delay significantly.

• Journal of Broadcast Engineering
• /
• v.26 no.1
• /
• pp.88-98
• /
• 2021

Recently, the crime prevention and culprit identification even in a low illuminated environment by CCTV is becoming ever more important. In a low lighting situation, CCTV applications capture images under infrared lighting since it is unobtrusive to human eye. Although the infrared lighting leads to advantage of capturing an image with abundant fine texture information, it is hard to capture the color information which is very essential in identifying certain objects or persons in CCTV images. In this paper, we propose a method to acquire color information through DCGAN from an image captured by CCTV in a low lighting environment with infrared lighting and a method to remove color noise in the acquired color image.

• Journal of Practical Engineering Education
• /
• v.13 no.3
• /
• pp.491-496
• /
• 2021

In this thesis, we intend to use CCTVs installed in existing parking lots to understand the current status of parking lots and provide real-time information to users through Android applications. It describes how to set the ROI in the parking area using YOLO V3 and how to provide the number of vacancies that change in real time through the set ROI, and describes how to link CCTV-server-user using IMAGE ZMQ and FIREBASE. The user can know the real-time situation of the parking lot near the destination before arriving through the application and can come up with various measures accordingly.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.6
• /
• pp.309-317
• /
• 2022

In this paper, we propose a machine learning-based copy-move forgery detection network with dual branches. Because the rotation or scaling operation is frequently involved in copy-move forger, the conventional convolutional neural network is not effectively applied in detecting copy-move tampering. Therefore, we divide the input into rotation-invariant and scaling-invariant features based on the wavelet coefficients. Each of the features is input to different branches having the same structure, and is fused in the combination module. Each branch comprises feature extraction, correlation, and mask decoder modules. In the proposed network, VGG16 is used for the feature extraction module. To check similarity of features generated by the feature extraction module, the conventional correlation module used. Finally, the mask decoder model is applied to develop a pixel-level localization map. We perform experiments on test dataset and compare the proposed method with state-of-the-art tampering localization methods. The results demonstrate that the proposed scheme outperforms the existing approaches.

• Smart Structures and Systems
• /
• v.29 no.4
• /
• pp.599-616
• /
• 2022

Bridge displacement contains vital information for bridge condition and performance. Due to the limits of direct displacement measurement methods, the indirect displacement reconstruction methods based on the strain or acceleration data are also developed in engineering applications. There are still some deficiencies of the displacement reconstruction methods based on strain or acceleration in practice. This paper proposed a novel method based on long short-term memory (LSTM) networks to reconstruct the bridge dynamic displacements with the strain and acceleration data source. The LSTM networks with three hidden layers are utilized to map the relationships between the measured responses and the bridge displacement. To achieve the data fusion, the input strain and acceleration data need to be preprocessed by normalization and then the corresponding dynamic displacement responses can be reconstructed by the LSTM networks. In the numerical simulation, the errors of the displacement reconstruction are below 9% for different load cases, and the proposed method is robust when the input strain and acceleration data contains additive noise. The hyper-parameter effect is analyzed and the displacement reconstruction accuracies of different machine learning methods are compared. For experimental verification, the errors are below 6% for the simply supported beam and continuous beam cases. Both the numerical and experimental results indicate that the proposed data fusion method can accurately reconstruct the displacement.