http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
김에덴,고석갑,손승철,이병탁,Kim, E.D.,Ko, S.K.,Son, S.C.,Lee, B.T. 한국전자통신연구원 2021 전자통신동향분석 Vol.36 No.4
Data imputation is a crucial issue in data analysis because quality data are highly correlated with the performance of AI models. Particularly, it is difficult to collect quality time-series data for uncertain situations (for example, electricity blackout, delays for network conditions). Thus, it is necessary to research effective methods of time-series data imputation. Many studies on time-series data imputation can be divided into 5 parts, including statistical based, matrix-based, regression-based, deep learning (RNN and GAN) based methodologies. This study reviews and organizes these methodologies. Recently, deep learning-based imputation methods are developed and show excellent performance. However, it is associated to some computational problems that make it difficult to use in real-time system. Thus, the direction of future work is to develop low computational but high-performance imputation methods for application in the real field.
레이저유도 플라즈마 분광법을 이용한 폐금속 분류를 위한 추정 연성정보 기반의 최빈 분류 기술
김에덴,장혜민,신성호,정성호,황의석,Kim, Eden,Jang, Hyemin,Shin, Sungho,Jeong, Sungho,Hwang, Euiseok 한국자원리싸이클링학회 2018 資源 리싸이클링 Vol.27 No.1
In this study, a novel soft information based most probable classification scheme is proposed for sorting recyclable metal alloys with laser induced breakdown spectroscopy (LIBS). Regression analysis with LIBS captured spectrums for estimating concentrations of common elements can be efficient for classifying unknown arbitrary metal alloys, even when that particular alloy is not included for training. Therefore, partial least square regression (PLSR) is employed in the proposed scheme, where spectrums of the certified reference materials (CRMs) are used for training. With the PLSR model, the concentrations of the test spectrum are estimated independently and are compared to those of CRMs for finding out the most probable class. Then, joint soft information can be obtained by assuming multi-variate normal (MVN) distribution, which enables to account the probability measure or a prior information and improves classification performance. For evaluating the proposed schemes, MVN soft information is evaluated based on PLSR of LIBS captured spectrums of 9 metal CRMs, and tested for classifying unknown metal alloys. Furthermore, the likelihood is evaluated with the radar chart to effectively visualize and search the most probable class among the candidates. By the leave-one-out cross validation tests, the proposed scheme is not only showing improved classification accuracies but also helpful for adaptive post-processing to correct the mis-classifications.
전력 데이터의 특징 추출 및 XGBoost를 이용한 숙박 업소 재실 여부 판단
김에덴 ( Eden Kim ),고석갑 ( Seok-gap Ko ),손승철 ( Seung-chul Son ),이형옥 ( Hyung-ok Lee ),이병탁 ( Byung-tak Lee ) 한국정보처리학회 2020 한국정보처리학회 학술대회논문집 Vol.27 No.1
스마트미터의 기술 발달과 보급으로 인해 전력데이터의 수집이 보다 수월 해짐에 따라 각 시스템에 효율적인 맞춤 서비스 제공을 위한 전력 데이터 분석 기술에 관한 다양한 연구가 활발하게 진행되고 있다. 관련하여 본 논문에서는 숙박업소의 각 방마다 전력소비량을 측정 및 수집하여 전력소비패턴을 분석하고 특징 추출 및 XGBoost 를 이용한 머신러닝 분석방법으로 각 방의 사람 재실 여부를 판별하는 방법을 소개한다. 이와 같은 연구를 통해 추후 숙박업소 혹은 숙박업소를 이용하는 소비자들의 맞춤 서비스 제공에 응용 및 적용 할 수 있다.
비대면 문장 따라 말하기에 나타난 3∼5세 아동의 구문 능력
김에덴 ( Kim E-den ),박미혜 ( Park Mee-hye ) 한국특수아동학회 2021 특수아동교육연구 Vol.23 No.4
Purpose: This study investigated syntactic development of children aged 3 to 5, and the results were analyzed after conducting sentence repetition tasks for all 45 children, 15 for each age. Method: The sentence repetition task consisted of 26 questions, which were adjusted according to the structure of the sentence (short, complex, conjunctive, and embedded sentence) and the length of the sentence(3, 4, 5-word). According to the results of the repetition tasks, one-way ANOVA was conducted to see if there were differences between age groups for the calculated scores. As a post-hoc test, the Tukey test was conducted. In addition, error type analysis was performed by age to examine the characteristics of syntactic development. Results: First, the sentence repetition scores of children aged 3 to 5 years increased with age. Second, the score according to sentence length the score increased according to age in the 4-word and 5-word tasks. 3-year-old children showed slightly higher scores for complex sentences than short sentences, but 4-year-old and 5-year-old showed higher scores for short sentences. Children of all ages showed lower scores for embedded sentences compared to conjunctive sentences. Third, as the results of analyzing the frequency of error types, it was found that the overall error frequency decreased significantly as the children age increased, and omissions were found to be the most frequent in all age groups regardless of sentence components. Conclusion: It is judged that non-face-to-face evaluation is worth using in the field. In addition, it is suggested that many systematic non-face-to-face studies should be conducted in various fields at a time when interest in non-face-to-face evaluation or rehabilitation is growing.
이준기,박상준,김낙우,김에덴,고석갑,J. Lee,S. Park,N.W. Kim,E. Kim,S.K. Ko 한국전자통신연구원 2024 전자통신동향분석 Vol.39 No.1
In natural language processing, large language models such as GPT-4 have recently been in the spotlight. The performance of natural language processing has advanced dramatically driven by an increase in the number of model parameters related to the number of acceptable input tokens and model size. Research on multimodal models that can simultaneously process natural language and image data is being actively conducted. Moreover, natural-language and image-based reasoning capabilities of large language models is being explored in robot artificial intelligence technology. We discuss research and related patent trends in robot task planning and code generation for robot control using large language models.
고석갑(SeokKap Ko),오승민(Seungmin Oh),김에덴(Eden Kim),이병탁(ByungTak Lee) 대한전자공학회 2020 대한전자공학회 학술대회 Vol.2020 No.8
일반적인 머신러닝 방식은 충분한 레이블 데이터를 이용해 모델을 학습시킨 후, 그 모델을 이용하여 예측을 수행하는 것이다. 만약 학습에 필요한 데이터가 추가로 생기는 경우, 모델을 재학습 시켜야 한다. 연속학습은 이렇게 학습해야 할 데이터가 계속 추가되는 경우 효율적으로 학습하는 과정을 말한다. 그런데 추가된 데이터를 이용해 모델을 재학습하는 경우 성능 저하가 발생할 수 있다. 그래서 기존 데이터에 추가 데이터를 합친 다음, 모델을 초기화하고 다시 학습 시켜야 한다. 이러한 과정은 많은 연산량과 학습 시간을 요구한다. 본 논문은, 이전 학습 단계 중 중간 정도 학습된 모델을 이용하여 추가학습을 진행하도록 하여, 성능과 학습 시간을 개선하는 방법을 제시한다. MNIST 데이터셋과 CNN을 이용한 실험을 통하여 제안하는 방법의 효용성을 확인하였다.