• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.11
• /
• pp.481-492
• /
• 2023

This study explores how to build a Korean dataset to extract information from text using generative large language models. In modern society, mixed information circulates rapidly, and effectively categorizing and extracting it is crucial to the decision-making process. However, there is still a lack of Korean datasets for training. To overcome this, this study attempts to extract information using text-based zero-shot learning using a generative large language model to build a purposeful Korean dataset. In this study, the language model is instructed to output the desired result through prompt engineering in the form of "system"-"instruction"-"source input"-"output format", and the dataset is built by utilizing the in-context learning characteristics of the language model through input sentences. We validate our approach by comparing the generated dataset with the existing benchmark dataset, and achieve 25.47% higher performance compared to the KLUE-RoBERTa-large model for the relation information extraction task. The results of this study are expected to contribute to AI research by showing the feasibility of extracting knowledge elements from Korean text. Furthermore, this methodology can be utilized for various fields and purposes, and has potential for building various Korean datasets.

• Journal of Broadcast Engineering
• /
• v.23 no.6
• /
• pp.760-767
• /
• 2018

Advanced driver assistance system(ADAS) is a major technique in the intelligent vehicle field. The techniques for ADAS can be separated in two classes, i.e., methods that directly control the movement of vehicle and that indirectly provide convenience to driver. In this paper, we propose a novel system that gives a visual assistance to driver by translating a night road image to a day road image. We use the black box images capturing the front road view of vehicle as inputs. The black box images are cropped into three parts and simultaneously translated into day images by the proposed image translation module. Then, the translated images are recollected to original size. The experimental result shows that the proposed method generates realistic images and outperforms the conventional algorithms.

• ETRI Journal
• /
• v.41 no.4
• /
• pp.415-425
• /
• 2019

Generative adversarial networks (GAN) have been successfully used in many image restoration tasks, including image denoising, super-resolution, and compression artifact reduction. By fully exploiting its characteristics, state-of-the-art image restoration techniques can be used to generate images with photorealistic details. However, there are many applications that require faithful rather than visually appealing image reconstruction, such as medical imaging, surveillance, and video coding. We found that previous GAN-training methods that used a loss function in the form of a weighted sum of fidelity and adversarial loss fails to reduce fidelity loss. This results in non-negligible degradation of the objective image quality, including peak signal-to-noise ratio. Our approach is to alternate between fidelity and adversarial loss in a way that the minimization of adversarial loss does not deteriorate the fidelity. Experimental results on compression-artifact reduction and super-resolution tasks show that the proposed method can perform faithful and photorealistic image restoration.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.5
• /
• pp.1590-1609
• /
• 2021

Laughter is one of the most important nonverbal sound that human generates. It is a means for expressing his emotions. The acoustic and contextual features of this specific sound are different from those of speech and many difficulties arise during their modeling process. During this work, we propose an audio laughter generation system based on unsupervised generative models: the autoencoder (AE) and its variants. This procedure is the association of three main sub-process, (1) the analysis which consist of extracting the log magnitude spectrogram from the laughter database, (2) the generative models training, (3) the synthesis stage which incorporate the involvement of an intermediate mechanism: the vocoder. To improve the synthesis quality, we suggest two hybrid models (LSTM-VAE, GRU-VAE and CNN-VAE) that combine the representation learning capacity of variational autoencoder (VAE) with the temporal modelling ability of a long short-term memory RNN (LSTM) and the CNN ability to learn invariant features. To figure out the performance of our proposed audio laughter generation process, objective evaluation (RMSE) and a perceptual audio quality test (listening test) were conducted. According to these evaluation metrics, we can show that the GRU-VAE outperforms the other VAE models.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.3
• /
• pp.109-114
• /
• 2021

GAN(Generative Adversarial Network) is an image generation model, which is composed of a generator network and a discriminator network, and generates an image similar to a real image. Since the image generated by the GAN should be similar to the actual image, a loss function is used to minimize the loss error of the generated image. However, there is a problem that the loss function of GAN degrades the quality of the image by making the learning to generate the image unstable. To solve this problem, this paper analyzes GAN-related studies and proposes an edge GAN(eGAN) using edge detection. As a result of the experiment, the eGAN model has improved performance over the existing GAN model.

• Smart Structures and Systems
• /
• v.31 no.5
• /
• pp.501-516
• /
• 2023

Dynamic irregularity and acceleration of bridges subjected to high-speed trains provide crucial information for comprehensive evaluation of the health state of under-track structures. This paper proposes a novel approach for real-time estimation of vertical track dynamic irregularity and bridge acceleration using deep generative adversarial network (GAN) and vibration data from in-service train. The vehicle-body and bogie acceleration responses are correlated with the two target variables by modeling train-bridge interaction (TBI) through least squares generative adversarial network (LSGAN). To realize supervised learning required in the present task, the conventional LSGAN is modified by implementing new loss function and linear activation function. The proposed approach can offer pointwise and accurate estimates of track dynamic irregularity and bridge acceleration, allowing frequent inspection of high-speed railway (HSR) bridges in an economical way. Thanks to its applicability in scenarios of high noise level and critical resonance condition, the proposed approach has a promising prospect in engineering applications.

• Journal of the Semiconductor & Display Technology
• /
• v.20 no.3
• /
• pp.131-135
• /
• 2021

Numbers of studies related to optimization of design of organic light emitting diodes(OLED) through machine learning are increasing. We propose the generative method of the image to assess the performance of the device combining with machine learning technique. Principle parameter regarding dark spot growth mechanism of the OLED can be the key factor to determine the long-time performance. Captured images from actual device and randomly generated images at specific time and initial pinhole state are fed into the deep neural network system. The simulation reinforced by the machine learning technique can predict the device parameters accurately and faster. Similarly, the inverse design using multiple layer perceptron(MLP) system can infer the initial degradation factors at manufacturing with given device parameter to feedback the design of manufacturing process.

• International Journal of Computer Science & Network Security
• /
• v.21 no.8
• /
• pp.127-136
• /
• 2021

Real-time human recognition is a challenging task, as the images are captured in an unconstrained environment with different poses, makeups, and styles. This limitation is addressed by generating several facial images with poses, makeup, and styles with a single reference image of a person using Generative Adversarial Networks (GAN). In this paper, we propose deep learning-based human recognition using integration of GAN and Spatial Domain Techniques. A novel concept of human recognition based on face depiction approach by generating several dissimilar face images from single reference face image using Domain Transfer Generative Adversarial Networks (DT-GAN) combined with feature extraction techniques such as Local Binary Pattern (LBP) and Histogram is deliberated. The Euclidean Distance (ED) is used in the matching section for comparison of features to test the performance of the method. A database of millions of people with a single reference face image per person, instead of multiple reference face images, is created and saved on the centralized server, which helps to reduce memory load on the centralized server. It is noticed that the recognition accuracy is 100% for smaller size datasets and a little less accuracy for larger size datasets and also, results are compared with present methods to show the superiority of proposed method.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.7
• /
• pp.1794-1806
• /
• 2023

This study presents a method for capturing photographs of users as input and converting them into 2D character animation sprites using a generative adversarial network-based artificial intelligence network. Traditionally, 2D character animations have been created by manually creating an entire sequence of sprite images, which incurs high development costs. To address this issue, this study proposes a technique that combines motion videos and sample 2D images. In the 2D sprite generation process that uses the proposed technique, a sequence of images is extracted from real-life images captured by the user, and these are combined with character images from within the game. Our research aims to leverage cutting-edge deep learning-based image manipulation techniques, such as the GAN-based motion transfer network (impersonator) and background noise removal (U² -Net), to generate a sequence of animation sprites from a single image. The proposed technique enables the creation of diverse animations and motions just one image. By utilizing these advancements, we focus on enhancing productivity in the game and animation industry through improved efficiency and streamlined production processes. By employing state-of-the-art techniques, our research enables the generation of 2D sprite images with various motions, offering significant potential for boosting productivity and creativity in the industry.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2022.05a
• /
• pp.61-61
• /
• 2022

최근 머신러닝 기술의 발달에 따라 이를 활용한 레이더 자료기반 강우예측기법이 활발히 개발되고 있다. 기존 머신러닝을 이용한 강우예측모델 개발 관련 연구는 주로 한 지역에 대해 수행되며, 데이터 기반으로 훈련되는 머신러닝 기법의 특성상 개발된 모델이 훈련된 지역에 대해서만 좋은 성능을 보인다는 한계점이 존재한다. 이러한 한계점을 해결하기 위해 사전 훈련된 모델을 이용하여 새로운 데이터에 대해 모델을 훈련하는 전이학습 기법 (transfer learning)을 적용하여 여러 유역에 대한 강우예측모델을 개발하고자 하였다. 본 연구에서는 사전 훈련된 강우예측 모델로 생성적 적대 신경망 기반 기법(Generative Adversarial Network, GAN)을 이용한 미래 강우예측모델을 사용하였다. 해당 모델은 기상청에서 제공된 2014년~2017년 여름의 레이더 이미지 자료를 이용하여 초단기, 단기 강우예측을 수행하도록 학습시켰으며, 2018년 레이더 이미지 자료를 이용한 단기강우예측 모의에서 좋은 성능을 보였다. 본 연구에서는 훈련된 모델을 이용해 새로운 댐 유역(안동댐, 충주댐)에 대한 강우예측모델을 개발하기 위해 여러 전이학습 기법을 적용하고, 그 결과를 비교하였다. 결과를 통해 새로운 데이터로 처음부터 훈련시킨 모델보다 전이학습 기법을 사용하였을 때 좋은 성능을 보이는 것을 확인하였으며, 이를 통해 여러 댐 유역에 대한 모델 개발 시 전이학습 기법이 효율적으로 적용될 수 있음을 확인하였다.