• International Journal of Internet, Broadcasting and Communication
• /
• v.16 no.2
• /
• pp.161-168
• /
• 2024

As the size of big data models grows, distributed training is emerging as an essential element for large-scale machine learning tasks. In this paper, we propose ParamHub for distributed data training. During the training process, this agent utilizes the provided data to adjust various conditions of the model's parameters, such as the model structure, learning algorithm, hyperparameters, and bias, aiming to minimize the error between the model's predictions and the actual values. Furthermore, it operates autonomously, collecting and updating data in a distributed environment, thereby reducing the burden of load balancing that occurs in a centralized system. And Through communication between agents, resource management and learning processes can be coordinated, enabling efficient management of distributed data and resources. This approach enhances the scalability and stability of distributed machine learning systems while providing flexibility to be applied in various learning environments.

• KIPS Transactions on Computer and Communication Systems
• /
• v.11 no.7
• /
• pp.205-216
• /
• 2022

One of the many tools used for distributed deep learning training is Kubeflow, which runs on Kubernetes, a container orchestration tool. TensorFlow jobs can be managed using the existing operator provided by Kubeflow. However, when considering the distributed deep learning training jobs based on the parameter server architecture, the scheduling policy used by the existing operator does not consider the task affinity of the distributed training job and does not provide the ability to dynamically allocate or release resources. This can lead to long job completion time and low resource utilization rate. Therefore, in this paper we proposes a new operator that efficiently schedules distributed deep learning training jobs to minimize the job completion time and increase resource utilization rate. We implemented the new operator by modifying the existing operator and conducted experiments to evaluate its performance. The experiment results showed that our scheduling policy improved the average job completion time reduction rate of up to 84% and average CPU utilization increase rate of up to 92%.

• 제어로봇시스템학회:학술대회논문집
• /
• 1996.10b
• /
• pp.1424-1427
• /
• 1996

OTS(Operator Training System) requires computation for the systematic training in real-time. So we have developed a distributed operator training system that is composed of workstation based server and PC based user modules. Sever and OM(OTS Manager) modules are located in the workstation server and user modules are located in PCs. User modules have DCS-like user interfaces and transfer data with OM over the coaxial ethernet. This paper delineates a total system architecture and definition of data transferring between OM and User module. Having applied this system to a batch process, we could analyze operator's tasks.

• Proceedings of the KSPS conference
• /
• 2006.05a
• /
• pp.37-40
• /
• 2006

In this paper, we propose a VQ codebook design of speech recognition feature parameters in order to improve the performance of a distributed speech recognition system. For the context-dependent HMMs, a VQ codebook should be correlated with phonetic distributions in the training data for HMMs. Thus, we focus on a selection method of training data based on phonetic distribution instead of using all the training data for an efficient VQ codebook design. From the speech recognition experiments using the Aurora 4 database, the distributed speech recognition system employing a VQ codebook designed by the proposed method reduced the word error rate (WER) by 10% when compared with that using a VQ codebook trained with the whole training data.

• Journal of the Korean Dietetic Association
• /
• v.17 no.4
• /
• pp.403-415
• /
• 2011

The present study was performed to evaluate the job training needs of school nutrition teachers in order to enhance their specialty. Three hundred and forty questionnaires were distributed to school nutrition teachers working at primary and high schools in the Gyeongbuk area while 45 were distributed to professors during 2010~2011. Three hundred and two questionnaires from school nutrition teachers and 33 from professors were returned and analyzed. The rate of teachers practicing nutrition education was 54%, and the educational content was obtained mainly from the internet. The top three problems the teachers encountered were 'lack of standardized educational materials', 'inexperience of teaching', and 'insufficiency of expert knowledge'. The teachers recognized 'training program' as the best solution. However, the job training program operated immediately after teachers were appointed scored only 3.03 out of 5.00. Important contents of the training program ranked highly by the teachers were 'development of education materials', 'nutrition counseling', and 'teaching method'. The professors included 'expert knowledge' in their top three contents. Both the teachers and professors agreed to increase the frequency of 'practice' in training methods. Other factors the teachers considered to be important were high quality, diversity, ability of the instructor, training cycle, and the institution in charge. From these results, it can be concluded that efficient job training programs are needed for school nutrition teachers according to the importance of the education contents and training methods. It is therefore suggested that a cooperation committee be composed of an educator, educatee, and related personal in a local education office in order to operate the program.

• The Journal of Bigdata
• /
• v.5 no.1
• /
• pp.189-195
• /
• 2020

We consider the stale problem which makes the training speed slow in the field of deep learning. The problem can be formulated as a single-machine scheduling problem with generalized due dates in which the objective is to minimize the total earliness and tardiness. We show that the problem can be solved in polynomial time if the orders of the small and the large jobs in an optimal schedule are known in advance.

• Journal of Practical Engineering Education
• /
• v.13 no.3
• /
• pp.453-460
• /
• 2021

The purpose of this study is to understand the educational types and educational status of institutions that operate virtual training contents distributed by K University's Online Lifelong Education Center and to examine the role and improvement measures of institutions that supply virtual training contents. To this end, a survey was conducted on 56 institutions that operate virtual training contents distributed by K University's Online Lifelong Education Center in 2020, and a survey of 44 institutions that responded finally was analyzed. The analysis results are as follows. First, virtual training education media responded that a total of 52 courses were applied to the content process using PC, the theory+virtual training type is the most, and one to two weeks of practice period were the most. In addition, when looking at the current status of education, the subjects of virtual training were mainly students, and the institution itself recruited, and the largest number of respondents was 1-20, the number of courses opened was 1-3, and the number of teachers and instructors in charge was 1-3. Second, they responded that it is necessary to develop various virtual training contents as a virtual training content supply institution, present ways to apply and utilize virtual training curriculum, and improve the quality of virtual training content. Third, as an improvement plan for virtual training content distribution institutions, it demanded strengthening the linkage of virtual training content, training using virtual training content by teachers and instructors, suggesting various virtual training content linkage, and strengthening virtual training content quality management. Therefore, this study is meaningful in that it identified the overall operation status and status of virtual training content management institutions and examined ways to establish and improve the role of public virtual training supply institutions.

• Journal of the Korean Society of Marine Environment & Safety
• /
• v.19 no.3
• /
• pp.270-276
• /
• 2013

The educational process and result of onboard training should be evaluated according to the 1995 Amendments to the International Convention on Standards of Training, Certification and Watch-keeping for seafarers(STCW), 1978. In particular, the revised Convention requires that a trainee's seagoing service must be recorded in each cadet's Training Record Book approved by the maritime administration responsible for the issuance of certificates of competency. Trainees for certification under regulation III/1 of the STCW Convention are required to complete an approved on-board training programme. The purpose of this paper is to understand the compliance of the education for an approved on-board training programme. The questionnaire was distributed among 110 cadets being trained on board the training ship of the maritime college of the Mokpo National Maritime University. In this study, we conducted the questionnaire survey which is related to the on-board training programme such as marine engineering; controlling the operation of the ship and care for persons on board; electrical, electronic and control engineering; etc. The survey revealed that onboard training program was normally satisfactory, however, lack of practical training tools and time have accounted for most of the reasons for dissatisfaction. Therefore, it is our goal to enhance the satisfactory value of onboard training education by analyzing the reason of the dissatisfaction.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.2-4
• /
• 2021

In a distributed machine learning system, sharing gradients was considered safe because it did not share original training data. However, recent studies found that malicious attacker could completely restore the original training data from shared gradients. Gradient Leakage Attack is a technique that restoring original training data by exploiting theses vulnerability. In this study, we present the image transformation method based on Discrete Cosine Transform to defend against the Gradient Leakage Attack on the federated learning setting, which training in local devices and sharing gradients to the server. Experiment shows that our image transformation method cannot be completely restored the original data from Gradient Leakage Attack.

• The Journal of Asian Finance, Economics and Business
• /
• v.7 no.11
• /
• pp.499-506
• /
• 2020

The study examines the influence of individual characteristics, training content, and manager support on the effectiveness of on-the-job (OJT) training in the banking and finance industry. A simple random sampling technique was used to select the samples. Questionnaires were distributed to respondents in order to obtain the data. Using cross-sectional data obtained from 396 respondents in Bank A in Malaysia, the multiple regression results show that self-efficacy, motivation to learn, training content, and manager support have positive influence on OJT training effectiveness. Among all these factors, manager support is very highly correlated with OJT training effectiveness. The findings have given fruitful insight of the crucial roles of OJT training in the respective bank, particularly to bring forward the roles of systematic design and implementation of OJT training. This study is not only expanding knowledge in OJT and training, but offers managers practical insights in developing good OJT training program by considering employees need, capabilities, skills and job requirement. Furthermore, this study also provides a valuable framework in identifying the effectiveness of OJT training program for certain jobs. Further discussion of the research findings and its implications to theoretical knowledge of training and managers are promised at the end of the article.