• Journal of Convergence for Information Technology
• /
• v.11 no.9
• /
• pp.28-34
• /
• 2021

In the case of object detection using deep learning, both accuracy and real-time are required. However, it is difficult to use a deep learning model that processes a large amount of data in a limited resource environment. To solve this problem, this paper proposes an object detection model for small embedded devices. Unlike the general detection model, the model size was minimized by using a structure in which the pre-trained feature extractor was removed. The structure of the model was designed by repeatedly stacking lightweight convolution blocks. In addition, the number of region proposals is greatly reduced to reduce detection overhead. The proposed model was trained and evaluated using the public dataset PASCAL VOC. For quantitative evaluation of the model, detection performance was measured with average precision used in the detection field. And the detection speed was measured in a Raspberry Pi similar to an actual embedded device. Through the experiment, we achieved improved accuracy and faster reasoning speed compared to the existing detection method.

• International conference on construction engineering and project management
• /
• 2013.01a
• /
• pp.359-366
• /
• 2013

Although many research efforts have been conducted to address the effect of crew members' work skills (e.g., technical and planning skills) on work performance (e.g., work duration and quality) in construction projects, the relationship between skill and performance has generated a great deal of controversy in the field of management (Inkpen and Crossan 1995). This controversy can lead to under- or over-estimations of the overall project schedule, and can make it difficult for project managers to implement appropriate managerial policies for enhancing project performance. To address this issue, the following aspects need to be considered: (a) work performances are determined not only by individual-level work skill but also by the group-level work skill affected by work team members, each member's role, and any working behavior pattern; (b) work planning has significant effects on to what extent work skill enhances performance; and (c) different types of activities in construction require different types of work, skill, and team composition. This research, therefore, develops a system dynamics (SD) model to analyze the effects of both individual-and group-level (i.e., multi-level) skill on performances by utilizing the advantages of SD in capturing a feedback process and state changes, especially in human factors (e.g., attitude, ability, and behavior). The model incorporates: (a) a multi-level skill evolution and relevant behavior development mechanism within a work group; (b) the interaction among work planning, a crew's skill-learning, skill manifestation, and performances; and (c) the different work characteristics of each activity. This model can be utilized to implement appropriate work planning (e.g., work scope and work schedule) and crew management policies (e.g., work team composition and decision of each worker's role) with an awareness of crew's skill and work performance. Understanding the different characteristics of each activity can also support project managers in applying strategic work planning and crew management for a corresponding activity, which may enhance each activity's performance, as well as the overall project performance.

• Journal of Advanced Navigation Technology
• /
• v.28 no.4
• /
• pp.472-479
• /
• 2024

Half of the world's total population lives in cities, continuous urbanization is progressing, and the urban population is expected to exceed two-thirds of the total population by 2050. To resolve this phenomenon, the Korean government is focusing on building a new urban air mobility (UAM) industrial ecosystem. Airlines are also part of the UAM industry ecosystem and are preparing to improve efficiency in safe operations, passenger safety, aircraft operation efficiency, and punctuality. This study performs demand forecasting using time series data on the number of daily passengers on Korean Air's Gimpo to Jeju route from 2019 to 2023. For this purpose, statistical and machine learning models such as SARIMA, Prophet, CatBoost, and Random Forest are applied. Methods for effectively capturing passenger demand patterns were evaluated through various models, and the machine learning-based Random Forest model showed the best prediction results. The research results will present an optimal model for accurate demand forecasting in the aviation industry and provide basic information needed for operational planning and resource allocation.

• Transactions on Semiconductor Engineering
• /
• v.2 no.4
• /
• pp.46-51
• /
• 2024

Deep Reinforcement Learning (DRL) has demonstrated human-level performance in sequential decision-making tasks and enables edge devices to adapt autonomously to unknown environments. However, implementing DRL adaptation remains challenging due to its massive data interactions and extensive DNN computations. Existing FPGA-based DRL accelerators focus solely on computation acceleration, leading to prolonged adaptation times. This paper proposes an energy-efficient FPGA accelerator tailored for fast online DRL adaptation, leveraging three key innovations: 1) A Heterogeneous Replay Buffer (HRB) that reduces training iterations by up to 90%, 2) Mixed-Precision Selective Re-Training (MP-SELRET) that decreases computations by 12% while replacing 27.2% of 32-bit floating-point operations with 16-bit fixed-point operations, 3) A Mixed-Precision Heterogeneous Architecture (MPHA) that maximizes resource utilization and boosts throughput by 39.8%. The proposed accelerator significantly enhances the efficiency and speed of DRL adaptation, addressing the limitations of traditional scratch trainingmethods.

• Journal of Engineering Education Research
• /
• v.20 no.4
• /
• pp.12-20
• /
• 2017

This study aims to suggest implications on develop and operate NCS curriculum by analyzing of differences in perception and educational needs of university students, graduates, human resource managers(HR managers) on NCS basic job skill. The respondents of survey are composed of 533 university students in K university, 730 graduates and 106 HR managers. The major findings are as follows: first, the results showed that the importances of all competencies recognized by HR managers was higher than those of university students and graduates. And graduates perceived their level of competencies as higher than those of university students. Second, the needs for self-development skill, technical skill, and problem-solving skill was the highest within the students and graduates, meanwhile, the needs for the mathematical skill, resource management skill, and the organization understanding skill were low. In contrast, the results of this study showed that the HR managers has the high needs for self-development skill, communication skill and problem-solving skill. Also, HR managers has the low needs for mathematical skill, resource management skill, information skill, technical skill, organizational understanding skill, and job ethics. Based on the above results, we proposed the necessity to develop and apply the NCS curriculum considering the education needs recognized by current students, graduates, and HR managers.

• Proceedings of the Technology Innovation Conference
• /
• 2003.02a
• /
• pp.173-190
• /
• 2003

Demand forecasting is essential in establishing national and corporate strategy as well as the management of their resource. We forecast demand for multi-generation product using discrete choice model combining diffusion model The discrete choice model generally captures consumers'valuation of the product's qualify in the framework of a cross-sectional analysis. We incorporate diffusion effects into a discrete choice model in order to capture the dynamics of demand for multi-generation products. As an empirical application, we forecast demand for worldwide DRAM (dynamic random access memory) and each of its generations from 1999 to 2005. In so doing, we use the method of 'Technological Forecasting'for DRAM Density and Price of the generations based on the Moore's law and learning by doing, respectively. Since we perform our analysis at the market level, we adopt the inversion routine in using the discrete choice model and find that our model performs well in explaining the current market situation, and also in forecasting new product diffusion in multi-generation product markets.

• Journal of KIISE
• /
• v.42 no.11
• /
• pp.1397-1403
• /
• 2015

A named entity(NE) dictionary is an important resource for the performance of NE recognition. However, it is not easy to construct a NE dictionary manually since human annotation is time consuming and labor-intensive. To save construction time and reduce human labor, we propose a semi-automatic system for the construction of a NE dictionary. The proposed system constructs a pseudo-document with Wiki-categories per NE class by using an active learning technique. Then, it calculates similarities between Wiki entries and pseudo-documents using the BM25 model, a well-known information retrieval model. Finally, it classifies each Wiki entry into NE classes based on similarities. In experiments with three different types of NE class sets, the proposed system showed high performance(macro-average F1-score of 0.9028 and micro-average F1-score 0.9554).

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.12
• /
• pp.2885-2890
• /
• 2014

Localization of a sensor network node using machine learning has been recently studied. It is easy for Support vector machines algorithm to implement in high level language enabling parallelism. Raspberrypi is a linux system which can be used as a sensor node. Pi can be used to construct IP based Hadoop clusters. In this paper, we realized Support vector machine using python language and built a sensor network cluster with 5 Pi's. We also established a Hadoop software framework to employ MapReduce mechanism. In our experiment, we implemented the test sensor network with a variety of parameters and examined based on proficiency, resource evaluation, and processing time. The experimentation showed that with more execution power and memory volume, Pi could be appropriate for a member node of the cluster, accomplishing precise classification for sensor localization using machine learning.

• Journal of Convergence for Information Technology
• /
• v.8 no.4
• /
• pp.205-211
• /
• 2018

The purpose of this study is to classify appropriate content types according to the cyber education contents needed for employees of health and welfare and to develop contents based on specific instructional design strategies for each type. We divided the four types of content into instructor-led type contents, storytelling type contents, practice type contents, and case presentation type contents based on previous research and existing health and welfare cyber education contents. For each content type, a macro design strategies and a micro design strategies were derived, and contents were developed in accordance with the design strategies. Data were collected from 150 employees of health and welfare to analyze learning satisfaction. The average of learning satisfaction was 4.42. The results of this study will provide theoretical background and practical implications for the design and development of cyber education contents in health and welfare areas.

• Proceedings of the Korean Society of Crop Science Conference
• /
• 2022.10a
• /
• pp.89-89
• /
• 2022

Soybean (Glycine max (L.) Merr.) is a representative food resource. To preserve the integrity of soybean, it is necessary to protect soybean yield and seed quality from threats of various pests and diseases. Riptortus pedestris is a well-known insect pest that causes the greatest loss of soybean yield in South Korea. This pest not only directly reduces yields but also causes disorders and diseases in plant growth. Unfortunately, no resistant soybean resources have been reported. Therefore, it is necessary to identify the distribution and movement of Riptortus pedestris at an early stage to reduce the damage caused by insect pests. Conventionally, the human eye has performed the diagnosis of agronomic traits related to pest outbreaks. However, due to human vision's subjectivity and impermanence, it is time-consuming, requires the assistance of specialists, and is labor-intensive. Therefore, the responses and behavior patterns of Riptortus pedestris to the scent of mixture R were visualized with a 3D model through the perspective of artificial intelligence. The movement patterns of Riptortus pedestris was analyzed by using time-series image data. In addition, classification was performed through visual analysis based on a deep learning model. In the object tracking, implemented using the YOLO series model, the path of the movement of pests shows a negative reaction to a mixture Rina video scene. As a result of 3D modeling using the x, y, and z-axis of the tracked objects, 80% of the subjects showed behavioral patterns consistent with the treatment of mixture R. In addition, these studies are being conducted in the soybean field and it will be possible to preserve the yield of soybeans through the application of a pest control platform to the early stage of soybeans.