http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Jongpil Lee,Juhan Nam IEEE Signal Processing Society 2017 IEEE signal processing letters Vol.24 No.8
<P>Music auto-tagging is often handled in a similar manner to image classification by regarding the two-dimensional audio spectrogram as image data. However, music auto-tagging is distinguished from image classification in that the tags are highly diverse and have different levels of abstraction. Considering this issue, we propose a convolutional neural networks (CNN)-based architecture that embraces multi-level and multi-scaled features. The architecture is trained in three steps. First, we conduct supervised feature learning to capture local audio features using a set of CNNs with different input sizes. Second, we extract audio features from each layer of the pretrained convolutional networks separately and aggregate them altogether giving a long audio clip. Finally, we put them into fully connected networks and make final predictions of the tags. Our experiments show that using the combination of multi-level and multi-scale features is highly effective in music auto-tagging and the proposed method outperforms the previous state-of-the-art methods on the MagnaTagATune dataset and the Million Song Dataset. We further show that the proposed architecture is useful in transfer learning.</P>
JongPil Kim(김종필),NooRee Huh(허누리) 한국교육평가학회 2007 교육평가연구 Vol.20 No.2
미국에서 실시되는 표준화 검사의 경우, 대부분 사전 검사 문항(pretest item)의 통계치를 이용하여 검사를 제작하기 때문에, 정확한 사전 검사 문항의 모수치를 추정하는 일은 매우 중요하다. 특히, 사전 검사 문항에 대한 정확한 문항반응이론(IRT) 모수 추정치를 얻고자 하는 다양한 연구가 진행되어져 왔는데, 최근 Stout와 그 동료들은 (2003) 사전 검사문항의 보다 정확한 추정치를 얻기 위해, 상관이 높은 다른 하나의 시험에서 얻은 정보를 이용한 이른바 부차적 정보 방법을 제시했다. 본 연구는 그들의 연구를 바탕으로 부차적 정보 방법이, 적은 수의 피험자들이 응답한 사전 검사 문항의 정확한 IRT 모수치 추정에 얼마나 유용한지를 알아보기 위해, 부차적 정보 방법을 사용하지 않았을 경우와의 비교 분석을 시도하였다. 이를 위해 하나의 표준화 검사에 속해 있는 두 개의 과학시험(상관계수 > .8)에 대한 응답 자료를 이용하였는데, 이 두 과학 시험은 동일한 시험실시 시간, 동일한 전체 문항 수 및 동일한 수의 사전 검사 문항으로 구성되어 있다. 세가지의 다른 IRT 모형, 세 가지의 다른 피험자 수, 두 개의 각각 다른 내용 영역이 두 가지의 척도화 방법과 함께 연구에서 고려되었다. 연구 결과, 이모수 일차원 IRT 모형과 이 모수 다차원 IRT 모형에 고정 척도화 방법을 적용했을 때에 한하여, 부차적 정보 방법은 모수치 추정에 있어 오차를 감소하는 효과가 있음이 발견되었다. When a new test form is constructed based on item response theory (IRT), a relatively large number of examinees is usually required to obtain an accurate estimation of item statistics for pretest items. However, using a large sample size is almost impossible in real situations. To increase the precision of pretest item estimation with a small sample size, some studies have investigated different approaches in item estimation. Recently, Stout and his colleagues (2003) explored the usefulness of a collateral information approach with IRT models in improving pretest item parameter estimation. This study expanded Stout et al."s study by comparing collateral information methods with a non-collateral information method considering several factors to investigate the effectiveness of using collateral information with IRT models in estimating pretest items with relatively small sample sizes. Real data sets obtained from two standardized science test administrations, which have the same testing time, test length, and the number of pretest items, were used for this study. In this study, three sample sizes (100, 200, and 400), two item parameter estimate methods (fixed and methods), two content areas (content P and G) were considered using three IRT models (the two and three parameter logistic unidimensional and the two parameter logistic multidimensional IRT models). The results of this study indicated that the use of collateral information with the 2-PL unidimensional and multidimensional IRT models under a fixed calibration method somewhat improved the parameter estimation of the pretest items. The collateral information approaches with a scaling method did not produce noticeable difference in parameter estimation even when the sample sizes were smaller. As a future direction, a simulation study is recommended to confirm the benefit of using collateral information with the 2-PL IRT model and the fixed method considering the properties of operational items based on various sample sizes.
Proposed Assessment for Quality of Experience of Live IPTV in Home Environments
Jongpil Jeong,Jae-Young Choi 한국인터넷방송통신학회 2015 Journal of Advanced Smart Convergence Vol.4 No.1
As the speed of networks that subscribers can use has greatly increased, demand for high-quality broadcast content, such as from Internet Protocol Television (IPTV) and Video on Demand (VoD), is likewise increasing. Therefore, while broadcasters are increasing content and channels, they are striving to improve consumer quality of experience (QoE) to differentiate themselves from competitors, including by producing higher physical-quality content. Recently, subjective measurement methods have been internationally standardized as the most reliable approach for measuring and evaluating IPTV QoE. However, a majority of these methods are performed in experimental environments and are based on the extremely brief viewing period of approximately ten seconds using original reference videos. It is actually difficult to apply standard evaluation methods based on a ten-second viewing interval to assess real broadcast watching of IPTV or other services that involve a longer time (i.e., more than thirty minutes). In this paper, we therefore propose a method that accommodates actual viewing environments. Using the mean opinion score, we experimentally analyze the effects of evaluation interval changes under actual conditions in which IPTV service is provided. In addition, we propose improvements by applying the results into actual live broadcast IPTV service and by analyzing consumer service QoE.
( Jongpil Cheon ),( Sungwon Chung ) 한국교육공학회 2014 한국교육공학회 학술대회발표자료집 Vol.2014 No.2
This research examined the effects of different types of stimuli (i.e., reflection and prediction) during pauses between segments in an animated instruction. In the first study, a total of 115 college students were randomly assigned to one of the six conditions: repetition, plain pause, passive reflection, active reflection, passive prediction, and active prediction. The second study was conducted to determine how shortened pause time affect the impact of stimuli types on learning performance and perceptions. The results showed that the reflection groups outperformed other groups on concept tests in the first study, while the prediction groups outperformed the plain pause group in the second study. Also, perceived usefulness of pause toward passive prediction was lower than other stimuli. The findings prove the potential role of stimulus during pause, however, the mixed results suggest that further studies should investigate the benefits of the two types of stimuli in various conditions.
High Efficiency Grid-Connected Multi String PV PCS using H-bridge Multi-level Topology
JongPil Lee,ByungDuk Min,TaeJin Kim,Honnyong Cha,DongWook Yoo,JiYoon Yoo 전력전자학회 2011 ICPE(ISPE)논문집 Vol.2011 No.5
In this paper, a novel topology is proposed that can significantly increase the efficiency of photovoltaic (PV) system. The multi level topology consists of several H-bridge cells connected in series, each one connected to a string of PV modules. The proposed topology offers advantages such as the operation at lower switching frequency or lower current ripple compared to conventional two level topologies. The control algorithm permits the independent control of each dc link voltage, enabling, in this way, the tracking of the maximum power point for each string of PV modules. The proposed topology is implemented for 240㎾ power conditioning system. The experimental results show that the proposed topology has good performance.