• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.447-451
• /
• 2022

본 논문에서는 사용자가 대화 텍스트 방식의 입력을 주었을 때 이를 키워드 중심으로 변환하여 이미지를 생성해내는 방식을 제안한다. 대화 텍스트란 채팅 등에서 주로 사용하는 형식의 구어체를 말하며 이러한 텍스트 형식은 텍스트 기반 이미지 생성 모델이 적절한 아웃풋 이미지를 생성하기 어렵게 만든다. 이를 해결하기 위해 대화 텍스트를 키워드 중심 텍스트로 바꾸어 텍스트 기반 이미지 생성 모델의 입력으로 변환하는 과정이 이미지 생성의 질을 높이는 좋은 방안이 될 수 있는데 이러한 태스크에 적합한 학습 데이터는 충분하지 않다. 본 논문에서는 이러한 문제를 다루기 위한 하나의 방안으로 사전학습된 초대형 언어모델인 KoGPT 모델을 활용하며, 퓨샷 러닝을 통해 적은 양의 직접 제작한 데이터만을 학습시켜 대화 텍스트 기반의 이미지 생성을 구현하는 방법을 제안한다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.105-111
• /
• 2021

한국어 사실 확인 과제는 학습 자료의 부재로 인해 연구에 어려움을 겪고 있다. 본 논문은 수작업으로 구성된 학습 자료를 토대로 자연어 생성 모델을 이용하여 한국어 사실 확인 자료를 구축하는 방법을 제안한다. 본 연구는 임의의 근거를 기반으로 하는 주장을 생성하는 방법 (E2C)과 임의의 주장을 기반으로 근거를 생성하는 방법 (C2E)을 모두 실험해보았다. 이때 기존 학습 자료에 위 두 학습 자료를 각각 추가하여 학습한 사실 확인 분류기가 기존의 학습 자료나 영문 사실 확인 자료 FEVER를 국문으로 기계 번역한 학습 자료를 토대로 구성된 분류기보다 평가 자료에 대해 높은 성능을 기록하였다. 또한, C2E 방법의 경우 수작업으로 구성된 자료 없이 기존의 자연어 추론 과제 자료와 HyperCLOVA Few Shot 예제만으로도 높은 성능을 기록하여, 비지도 학습 방식으로 사실 확인 자료를 구축할 수 있는 가능성 역시 확인하였다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.396-401
• /
• 2021

초거대 언어모델를 활용한 퓨샷(few shot) 학습법은 여러 자연어 처리 문제에서 좋은 성능을 보였다. 하지만 데이터를 활용한 추가 학습으로 문제를 추론하는 것이 아니라, 이산적인 공간에서 퓨샷 구성을 통해 문제를 정의하는 방식은 성능 향상에 한계가 존재한다. 이를 해결하기 위해 초거대 언어모델의 모수 전체가 아닌 일부를 추가 학습하거나 다른 신경망을 덧붙여 연속적인 공간에서 추론하는 P-tuning과 같은 데이터 기반 추가 학습 방법들이 등장하였다. 본 논문에서는 문맥에 따른 질의 정규화 문제를 대화형 음성 검색 서비스에 맞게 직접 정의하였고, 초거대 언어모델을 P-tuning으로 추가 학습한 경우 퓨샷 학습법 대비 정확도가 상승함을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.171-176
• /
• 2023

최근 등장한 대규모 데이터로 사전학습된 자연어 생성 모델들은 대화 능력 및 코드 생성 태스크등에서 인상적인 성능을 보여주고 있어, 본 논문에서는 대형 언어 모델 (LLM)의 한국어 질문을 SQL 쿼리 (Text-to-SQL) 변환하는 성능을 평가하고자 한다. 먼저, 영어 Text-to-SQL 벤치마크 데이터셋을 활용하여 영어 질의문을 한국어 질의문으로 번역하여 한국어 Text-to-SQL 데이터셋으로 만들었다. 대형 생성형 모델 (GPT-3 davinci, GPT-3 turbo) 의 few-shot 세팅에서 성능 평가를 진행하며, fine-tuning 없이도 대형 언어 모델들의 경쟁력있는 한국어 Text-to-SQL 변환 성능을 확인한다. 또한, 에러 분석을 수행하여 한국어 문장을 데이터베이스 쿼리문으로 변환하는 과정에서 발생하는 다양한 문제와 프롬프트 기법을 활용한 가능한 해결책을 제시한다.

• Proceedings of the Korea Contents Association Conference
• /
• 2009.05a
• /
• pp.263-267
• /
• 2009

With the increase of weight and influence of TV drama, Many researchers began to publish papers on TV drama. However, Those papers were mainly about social meaning of TV drama and viewing motives and attitudes of TV audience. However, only a few limited papers were about television image production techniques. The purpose of this thesis is to analyse the relationship between TV genre and image production technique with the data of top 100 of TNS Media Korea from the year 2004 to 2008. This paper found out that history drama uses more closeup, longshot, tracking and less waist shot than other genres, with shorter shot duration and longer scene duration.

• Journal of the Institute of Convergence Signal Processing
• /
• v.5 no.4
• /
• pp.294-300
• /
• 2004

According to the development of digital media technologies various algorithms for video clip matching have been proposed to match the video sequences efficiently. A large number of video search methods have focused on frame-wise query, whereas a relatively few algorithms have been presented for video clip matching or video shot matching. In this paper, we propose an efficient algorithm to index the video sequences and to retrieve the sequences for video clip query. To improve the accuracy and performance of video sequence matching, we employ the Cauchy function as a similarity measure between histograms of consecutive frames, which yields a high performance compared with conventional measures. The key frames extracted from segmented video shots can be used not only for video shot clustering but also for video sequence matching or browsing, where the key frame is defined by the frame that is significantly different from the previous frames. Experimental results with color video sequences show that the proposed method yields the high matching performance and accuracy with a low computational load compared with conventional algorithms.

• The KIPS Transactions:PartA
• /
• v.9A no.1
• /
• pp.29-36
• /
• 2002

For the efficient collaboration awareness, the CSCW (Computer-Supported Cooperative Work) systems provide functions such as telepointing, snap-shot, and document/project tracking. Among these, telepointing allows an event occurrence on the shared window of a local computing system to be presented in remote computing systems. Although telepointing has been mentioned in the literature, few research papers deal with the functional requirements and protocols for telepointing. This parer describes the functional requirements and payload format for RTP (Real-Time Transport Protocol)-based telepointing. In order to evaluate the performance of the proposed RTP-barred telepointing, we have implemented a whiteboard and compared it with that of the Microsoft Netmeeting. The evaluation results show that the proposed telepointing performs better in terms of CPU usage and the amount of network traffic.

• The Bulletin of The Korean Astronomical Society
• /
• v.44 no.1
• /
• pp.54.2-54.2
• /
• 2019

Removing noise which occurs inevitably when taking image data has been a big concern. There is a way to raise signal-to-noise ratio and it is regarded as the only way, image stacking. Image stacking is averaging or just adding all pixel values of multiple pictures taken of a specific area. Its performance and reliability are unquestioned, but its weaknesses are also evident. Object with fast proper motion can be vanished, and most of all, it takes too long time. So if we can handle single shot image well and achieve similar performance, we can overcome those weaknesses. Recent developments in deep learning have enabled things that were not possible with former algorithm-based programming. One of the things is generating data with more information from data with less information. As a part of that, we reproduced stacked image from single shot image using a kind of deep learning, conditional generative adversarial network (cGAN). r-band camcol2 south data were used from SDSS Stripe 82 data. From all fields, image data which is stacked with only 22 individual images and, as a pair of stacked image, single pass data which were included in all stacked image were used. All used fields are cut in $128{\times}128$ pixel size, so total number of image is 17930. 14234 pairs of all images were used for training cGAN and 3696 pairs were used for verify the result. As a result, RMS error of pixel values between generated data from the best condition and target data were $7.67{\times}10^{-4}$ compared to original input data, $1.24{\times}10^{-3}$. We also applied to a few test galaxy images and generated images were similar to stacked images qualitatively compared to other de-noising methods. In addition, with photometry, The number count of stacked-cGAN matched sources is larger than that of single pass-stacked one, especially for fainter objects. Also, magnitude completeness became better in fainter objects. With this work, it is possible to observe reliably 1 magnitude fainter object.

• Journal of Astronomy and Space Sciences
• /
• v.31 no.3
• /
• pp.225-233
• /
• 2014

The first Korean satellite laser ranging (SLR) system, Daedeok SLR station (DAEK station) was developed by Korea Astronomy and Space Science Institute (KASI) in 2012, whose main objectives are space geodesy researches. In consequence, Korea became the $25^{th}$ country that operates SLR system supplementing the international laser tracking network. The DAEK station is designed to be capable of 2 kHz laser ranging with precision of a few mm both in daytime and nighttime observation of satellites with laser retro-reflector array (LRA) up to the altitude of 25,000 km. In this study, characteristics and specifications of DAEK station are investigated and its data quality is evaluated and compared with International Laser Ranging Service (ILRS) stations in terms of single-shot ranging precision. The analysis results demonstrated that the DAEK station shows good ranging performance to a few mm precision. Currently, the DAEK station is under normal operations at KASI headquarters, however, it will be moved to Sejong city in 2014 to function as a fundamental station for space geodesy researches in combination with other space geodesy systems (GNSS, VLBI, DORIS, etc.).

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.3
• /
• pp.148-155
• /
• 2024

Despite the increasing importance of irony and sarcasm detection in the field of natural language processing, research on the Korean language is relatively scarce compared to other languages. This study aims to experiment with various models for irony detection in Korean text. The study conducted irony detection experiments using KoBERT, a BERT-based model, and ChatGPT. For KoBERT, two methods of additional training on sentiment data were applied (Transfer Learning and MultiTask Learning). Additionally, for ChatGPT, the Few-Shot Learning technique was applied by increasing the number of example sentences entered as prompts. The results of the experiments showed that the Transfer Learning and MultiTask Learning models, which were trained with additional sentiment data, outperformed the baseline model without additional sentiment data. On the other hand, ChatGPT exhibited significantly lower performance compared to KoBERT, and increasing the number of example sentences did not lead to a noticeable improvement in performance. In conclusion, this study suggests that a model based on KoBERT is more suitable for irony detection than ChatGPT, and it highlights the potential contribution of additional training on sentiment data to improve irony detection performance.