• 말소리와 음성과학
• /
• 제12권2호
• /
• pp.29-37
• /
• 2020

최근 신경망 기반 심층학습 알고리즘의 적용으로 고전적인 Gaussian mixture model based hidden Markov model (GMM-HMM) 음성인식기에 비해 성능이 비약적으로 향상되었다. 또한 심층학습 기법의 장점을 더욱 잘 활용하는 방법으로 언어모델링 및 디코딩 과정을 통합처리 하는 종단간 음성인식 시스템에 대한 연구가 매우 활발히 진행되고 있다. 일반적으로 종단간 음성인식 시스템은 어텐션을 사용한 여러 층의 인코더-디코더 구조로 이루어져 있다. 때문에 종단간 음성인식 시스템이 충분히 좋은 성능을 내기 위해서는 많은 양의 음성과 문자열이 함께 있는 데이터가 필요하다. 음성-문자열 짝 데이터를 구하기 위해서는 사람의 노동력과 시간이 많이 필요하여 종단간 음성인식기를 구축하는 데 있어서 높은 장벽이 되고 있다. 그렇기에 비교적 적은 양의 음성-문자열 짝 데이터를 이용하여 종단간 음성인식기의 성능을 향상하는 선행연구들이 있으나, 음성 단일 데이터나 문자열 단일 데이터 한쪽만을 활용하여 진행된 연구가 대부분이다. 본 연구에서는 음성 또는 문자열 단일 데이터를 함께 이용하여 종단간 음성인식기가 다른 도메인의 말뭉치에서도 좋은 성능을 낼 수 있도록 하는 준교사 학습 방식을 제안했으며, 성격이 다른 도메인에 적응하여 제안된 방식이 효과적으로 동작하는지 확인하였다. 그 결과로 제안된 방식이 타깃 도메인에서 좋은 성능을 보임과 동시에 소스 도메인에서도 크게 열화되지 않는 성능을 보임을 알 수 있었다.

• 스마트미디어저널
• /
• 제11권2호
• /
• pp.9-17
• /
• 2022

본 연구는 야구 경기에서 선발 투수를 위기 상황 이전에 교체하기 위한 의사결정을 지원하는 예측 모델 구현을 목적으로 한다. 이를 위해 베이스볼 서번트(Baseball Savant)에서 제공하는 메이저리그 스탯캐스트 데이터를 활용하여, 선발 투수를 위기 상황 이전에 선제적으로 교체하는 예측 모델을 구현한다. 이를 위해 첫째, 데이터 탐색을 통해 선발 투수가 경기에서 직면하는 위기 상황을 도출하였다. 둘째, 선발 투수가 이닝 종료 전에 교체된 경우, 이전 이닝에서 교체하는 것으로 레이블을 구성하여 학습을 진행하였다. 학습된 모델을 비교한 결과 앙상블 기법을 기반으로 한 모델이 F1-Score가 65%로 가장 높은 예측 성능을 보였다. 본 연구의 실무적 의의는 제안하는 모델을 통해 선발 투수를 위기 상황 이전에 교체하여 팀의 승리 확률을 높이는 데 기여할 수 있으며, 경기 중 감독은 데이터 기반의 전략적 의사결정 지원을 받을 수 있을 것이다.

• 로봇학회논문지
• /
• 제14권1호
• /
• pp.40-49
• /
• 2019

Reinforcement learning has been applied to various problems in robotics. However, it was still hard to train complex robotic manipulation tasks since there is a few models which can be applicable to general tasks. Such general models require a lot of training episodes. In these reasons, deep neural networks which have shown to be good function approximators have not been actively used for robot manipulation task. Recently, some of these challenges are solved by a set of methods, such as Guided Policy Search, which guide or limit search directions while training of a deep neural network based policy model. These frameworks are already applied to a humanoid robot, PR2. However, in robotics, it is not trivial to adjust existing algorithms designed for one robot to another robot. In this paper, we present our implementation of Guided Policy Search to the robotic arms of the Baxter Research Robot. To meet the goals and needs of the project, we build on an existing implementation of Baxter Agent class for the Guided Policy Search algorithm code using the built-in Python interface. This work is expected to play an important role in popularizing robot manipulation reinforcement learning methods on cost-effective robot platforms.

• Safety and Health at Work
• /
• 제10권1호
• /
• pp.67-74
• /
• 2019

Background: Few studies have examined the effects of a forward rotating shift pattern on police employee performance and well-being. This study sought to compare sleep duration, cognitive performance, and vigilance at the start and end of each shift within a three-shift, forward rotating shift pattern, common in United Kingdom police forces. Methods: Twenty-three police employee participants were recruited from North Yorkshire Police (mean age, 43 years). The participants were all working the same, 10-day, forward rotating shift pattern. No other exclusion criteria were stipulated. Sleep data were gathered using both actigraphy and self-reported methods; cognitive performance and vigilance were assessed using a customized test battery, comprising five tests: motor praxis task, visual object learning task, NBACK, digital symbol substitution task, and psychomotor vigilance test. Statistical comparisons were conducted, taking into account the shift type, shift number, and the start and end of each shift worked. Results: Sleep duration was found to be significantly reduced after night shifts. Results showed a significant main effect of shift type in the visual object learning task and NBACK task and also a significant main effect of start/end in the digital symbol substitution task, along with a number of significant interactions. Conclusion: The results of the tests indicated that learning and practice effects may have an effect on results of some of the tests. However, it is also possible that due to the fast rotating nature of the shift pattern, participants did not adjust to any particular shift; hence, their performance in the cognitive and vigilance tests did not suffer significantly as a result of this particular shift pattern.

• 한국정보과학회 언어공학연구회:학술대회논문집(한글 및 한국어 정보처리)
• /
• 한국정보과학회언어공학연구회 2019년도 제31회 한글 및 한국어 정보처리 학술대회
• /
• pp.236-238
• /
• 2019

신경망 기계 번역(Neural Machine Translation)은 주로 지도 학습(Supervised learning)을 이용한 End-to-end 방식의 연구가 이루어지고 있다. 그러나 지도 학습 방법은 데이터가 부족한 경우에는 낮은 성능을 보이기 때문에 BERT와 같은 대량의 단일 언어 데이터로 사전학습(Pre-training)을 한 후에 미세조정(Finetuning)을 하는 Transfer learning 방법이 자연어 처리 분야에서 주로 연구되고 있다. 최근에 발표된 MASS 모델은 언어 생성 작업을 위한 사전학습 방법을 통해 기계 번역과 문서 요약에서 높은 성능을 보였다. 본 논문에서는 영어-한국어 기계 번역 성능 향상을 위해 MASS 모델을 신경망 기계 번역에 적용하였다. 실험 결과 MASS 모델을 이용한 영어-한국어 기계 번역 모델의 성능이 기존 모델들보다 좋은 성능을 보였다.

• 인터넷정보학회논문지
• /
• 제23권2호
• /
• pp.45-52
• /
• 2022

With the development of artificial intelligence analysis methods, especially machine learning, various fields are widely expanding their application ranges. However, in the case of classical music, there still remain some difficulties in applying machine learning techniques. Genre classification or music recommendation systems generated by deep learning algorithms are actively used in general music, but not in classical music. In this paper, we attempted to classify opera among classical music. To this end, an experiment was conducted to determine which criteria are most suitable among, composer, period of composition, and emotional atmosphere, which are the basic features of music. To generate emotional labels, we adopted zero-shot classification with four basic emotions, 'happiness', 'sadness', 'anger', and 'fear.' After embedding the opera libretto with the doc2vec processing model, the optimal number of clusters is computed based on the result of the elbow method. Decided four centroids are then adopted in k-means clustering to classify unsupervised libretto datasets. We were able to get optimized clustering based on the result of adjusted rand index scores. With these results, we compared them with notated variables of music. As a result, it was confirmed that the four clusterings calculated by machine after training were most similar to the grouping result by period. Additionally, we were able to verify that the emotional similarity between composer and period did not appear significantly. At the end of the study, by knowing the period is the right criteria, we hope that it makes easier for music listeners to find music that suits their tastes.

• 한국콘텐츠학회논문지
• /
• 제22권6호
• /
• pp.413-427
• /
• 2022

본 연구의 목적은 플립드 러닝과 프로젝트 기반 학습을 결합한 메타버스 게임화 교수법으로 대학생들의 과제가치와 학업적 자기효능감을 향상시키고 이를 검증하는 것이다. 본 연구의 대상은 청주의 K대학교에서 상담 심리학을 전공하는 16명의 대학생들이다. 대학생들은 각각 실험집단과 비교집단에 배정되었다. 실험집단에는 플립드 러닝과 프로젝트 기반 학습을 결합한 메타버스 게임화 교수법이 적용되었고 비교집단에는 단순 강의식 교수법이 적용되었다. 본 연구에서의 종속변인은 과제가치와 학업적 자기효능감이었다. 그리고 각 변인들을 사전, 사후에 측정하였다. 연구 결과 사후 검사에서 실험집단의 과제가치와 학업적 자기효능감은 비교집단에 비해 통계적으로 유의미하게 높았다. 본 연구의 결과는 플립드 러닝과 프로젝트 기반 학습을 결합한 메타버스 게임화 교수법이 대학생들의 과제가치 및 학업적 자기효능감의 향상에 효과적임을 시사한다.

• 공학교육연구
• /
• 제25권4호
• /
• pp.35-41
• /
• 2022

In order to enhance basic competencies for freshmen at engineering college, whose learning ability is gradually declining, a new course was developed to review basic mathematics and physics through a process of collecting opinions from fellow professors. Tests in six fields of math and physics with the same problems showed the correct answer rate rose from 24.8% at the beginning of the semester to 59.0% at the end of the semester after operating the course developed. According to the survey, the students' self-evaluated confidence on the basic competencies in 16 fields of math and physics showed a significant increase. Students with high confidence in basic competencies also received high actual grades. General high school graduates' confidence point in basic competencies improved from 54.7 at the beginning to 75.3 points at the end of the semester, while specialized high school graduates' enhanced from 38.3 to 64.0 which is higher than that of general high school graduates at the beginning of the semester.

• Asia pacific journal of information systems
• /
• 제17권4호
• /
• pp.61-95
• /
• 2007

Small-and-medium sized enterprises(SMEs) represent quite a large proportion of the industry as a whole in terms of the number of enterprises or employees. However researches on information system so far have focused on large companies, probably because SMEs were not so active in introducing information systems as larger enterprises. SMEs are now increasingly bringing in information systems such as ERP(Enterprise Resource Planning Systems) and some of the companies already entered the stage of ongoing use. Accordingly, researches should deal with the use of information systems by SME s operating under different conditions from large companies. This study examined factors and mechanism inducing faithful appropriation of information systems, in particular integrative systems such as ERP, in view of individuals` active feedback-seeking behavior. There are three factors expected to affect end users` feedback-seeking behavior for faithful appropriation of information systems. They are management support, peer IT champ support, and IT staff support. The main focus of the study is on how these factors affect feedback-seeking behavior and whether the feedback-seeking behavior plays the role of mediator for realizing faithful appropriation of information systems by end users. To examine the research model and the hypotheses, this study employed an empirical method based on a field survey. The survey used measurements mostly employed and verified by previous researches, while some of the measurements had gone through minor modifications for the purpose of the study. The survey respondents are individual employees of SMEs that have been using ERP for one year or longer. To prevent common method bias, Task-Technology Fit items used as the control variable were made to be answered by different respondents. In total, 127 pairs of valid questionnaires were collected and used for the analysis. The PLS(Partial Least Squares) approach to structural equation modeling(PLS-Graph v.3.0) was used as our data analysis strategy because of its ability to model both formative and reflective latent constructs under small-and medium-size samples. The analysis shows Reliability, Construct Validity and Discriminant Validity are appropriate. The path analysis results are as follows; first, the more there is peer IT champ support, the more the end user is likely to show feedback-seeking behavior(path-coefficient=0.230, t=2.28, p<0.05). In other words, if colleagues proficient in information system use recognize the importance of their help, pass on what they have found to be an effective way of using the system or correct others' misuse, ordinary end users will be able to seek feedback on the faithfulness of their appropriation of information system without hesitation, because they know the convenience of getting help. Second, management support encourages ordinary end users to seek more feedback(path-coefficient=0.271, t=3.06, p<0.01) by affecting the end users' perceived value of feedback(path-coefficient=0.401, t=6.01, p<0.01). Management support is far more influential than other factors that when the management of an SME well understands the benefit of ERP, promotes its faithful appropriation and pays attention to employees' satisfaction with the system, employees will make deliberate efforts for faithful appropriation of the system. However, the third factor, IT staff support was found not to be conducive to feedback-seeking behavior from end users(path-coefficient=0.174, t=1.83). This is partly attributable to the fundamental reason that there is little support for end users from IT staff in SMEs. Even when IT staff provides support, end users may find it less important than that from coworkers more familiar with the end users' job. Meanwhile, the more end users seek feedback and attempt to find ways of faithful appropriation of information systems, the more likely the users will be able to deploy the system according to the purpose the system was originally meant for(path-coefficient=0.35, t=2.88, p<0.01). Finally, the mediation effect analysis confirmed the mediation effect of feedback-seeking behavior. By confirming the mediation effect of feedback-seeking behavior, this study draws attention to the importance of feedback-seeking behavior that has long been overlooked in research about information system use. This study also explores the factors that promote feedback-seeking behavior which in result could affect end user`s faithful appropriation of information systems. In addition, this study provides insight about which inducements or resources SMEs should offer to promote individual users' feedback-seeking behavior when formal and sufficient support from IT staff or an outside information system provider is hardly expected. As the study results show, under the business environment of SMEs, help from skilled colleagues and the management plays a critical role. Therefore, SMEs should seriously consider how to utilize skilled peer information system users, while the management should pay keen attention to end users and support them to make the most of information systems.

• 조명전기설비학회논문지
• /
• 제13권2호
• /
• pp.55-62
• /
• 1999

자동화 설비시스템에서 부품의 조립, 가공 등 복잡하고 정교한 임무를 수행시키기위해서는 end-effector의 이동경로 궤적에 대한 추적제어 뿐만 아니라 목표물에 대하여 접촉하는 힘의 궤적에 대한 추적제어가 필수적이다. 본 논문에서는 외란이나 시스템의 파라미터 변동 및 불확실설 등이 존재하는 자동화 설비시스템을 강인하고 정밀하게 제어할 수 ldT도로 하기 위해 동적 신경망 처리(DNP)라 불리우는 신경망 제어기를 설계하였다. 또한 자동화 설비시스템의 매니플레이터에 역기구학적인 좌표변환을 계산하기 위한 학습구조를 개발하였으며, DNP가 이용될수 있는 예를 설명하고자 한다. 제안된 동적 신경망인 DNP의 구조와 학습 알고리즘을 제시하고 컴퓨터 모의 실험을 통해 DNP를 이용한 제안된 학습법의 성능을 확인한다.