• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.89-100
• /
• 2023

Recently, on-device artificial intelligence (AI) solutions using mobile devices and embedded edge devices have emerged in various fields, such as computer vision, to address network traffic burdens, low-energy operations, and security problems. Although vision transformer deep learning models have outperformed conventional convolutional neural network (CNN) models in computer vision, they require more computations and parameters than CNN models. Thus, they are not directly applicable to embedded edge devices with limited hardware resources. Many researchers have proposed various model compression methods or lightweight architectures for vision transformers; however, there are only a few studies evaluating the effects of model compression techniques of vision transformers on performance. Regarding this problem, this paper presents a performance evaluation of vision transformers on embedded platforms. We investigated the behaviors of three vision transformers: DeiT, LeViT, and MobileViT. Each model performance was evaluated by accuracy and inference time on edge devices using the ImageNet dataset. We assessed the effects of the quantization method applied to the models on latency enhancement and accuracy degradation by profiling the proportion of response time occupied by major operations. In addition, we evaluated the performance of each model on GPU and EdgeTPU-based edge devices. In our experimental results, LeViT showed the best performance in CPU-based edge devices, and DeiT-small showed the highest performance improvement in GPU-based edge devices. In addition, only MobileViT models showed performance improvement on EdgeTPU. Summarizing the analysis results through profiling, the degree of performance improvement of each vision transformer model was highly dependent on the proportion of parts that could be optimized in the target edge device. In summary, to apply vision transformers to on-device AI solutions, either proper operation composition and optimizations specific to target edge devices must be considered.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.12
• /
• pp.1-7
• /
• 2023

The advent of deep learning technologies has led to the development of various medical applications, making healthcare services more convenient and effective. Among these applications, heart rate estimation is considered a vital method for assessing an individual's health. Traditional methods, such as photoplethysmography through smart watches, have been widely used but are invasive and require additional hardware. Recent advancements allow for contactless heart rate estimation through facial image analysis, providing a more hygienic and convenient approach. In this paper, we propose a lightweight methodology capable of accurately estimating heart rate in mobile environments, using a specialized 2-channel network structure based on 2D convolution. Our method considers both subtle facial movements and color changes resulting from blood flow and muscle contractions. The approach comprises two major components: an Encoder for analyzing image features and a regression layer for evaluating Blood Volume Pulse. By incorporating both features simultaneously our methodology delivers more accurate results even in computing environments with limited resources. The proposed approach is expected to offer a more efficient way to monitor heart rate without invasive technology, particularly well-suited for mobile devices.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.11
• /
• pp.13-18
• /
• 2022

In this paper, we proposed VW-FedAVG(Validation based Weighted FedAVG) which updates the global model by weighting according to performance verification from the models of each device participating in the training. The first method is designed to validate each local client model through validation dataset before updating the global model with a server side validation structure. The second is a client-side validation structure, which is designed in such a way that the validation data set is evenly distributed to each client and the global model is after validation. MNIST, CIFAR-10 is used, and the IID, Non-IID distribution for image classification obtained higher accuracy than previous studies.

• Journal of Internet Computing and Services
• /
• v.24 no.5
• /
• pp.107-120
• /
• 2023

The public bicycle of Seoul, Ttareungyi, was launched at October 2015 to reduce traffic and carbon emissions in downtown Seoul and now, 2023 Oct, the cumulative number of user is upto 4 million and the number of bike is about 43,000 with about 2700 stations. However, super growth of Ttareungyi has caused the several problems, especially demand/supply mismatch, and thus the Seoul citizen has been complained about out of stock. In this point, this study conducted a real time demand forecasting model to prevent stock out bike at stations. To develop the model, the research team gathered the rental·return transaction data of 20,000 bikes in whole 1600 stations for 2019 year and then analyzed bike usage, user behavior, bike stations, and so on. The forecasting model using machine learning is developed to predict the amount of rental/return on each bike station every hour through daily learning with the recent 90 days data with the weather information. The model is validated with MAE and RMSE of bike stations, and tested as a prototype service on the Seoul Bike Management System(Mobile App) for the relocation team of Seoul City.

• Tunnel and Underground Space
• /
• v.30 no.5
• /
• pp.462-472
• /
• 2020

Rock classification is fundamental discipline of exploring geological and geotechnical features in a site, which, however, may not be easy works because of high diversity of rock shape and color according to its origin, geological history and so on. With the great success of convolutional neural networks (CNN) in many different image-based classification tasks, there has been increasing interest in taking advantage of CNN to classify geological material. In this study, a feasibility of the deep CNN is investigated for automatically and accurately identifying rock types, focusing on the condition of various shapes and colors even in the same rock type. It can be further developed to a mobile application for assisting geologist in classifying rocks in fieldwork. The structure of CNN model used in this study is based on a deep residual neural network (ResNet), which is an ultra-deep CNN using in object detection and classification. The proposed CNN was trained on 10 typical rock types with an overall accuracy of 84% on the test set. The result demonstrates that the proposed approach is not only able to classify rock type using images, but also represents an improvement as taking highly diverse rock image dataset as input.

• Knowledge Management Research
• /
• v.18 no.1
• /
• pp.83-116
• /
• 2017

The purpose of this study is to analyze empirically analyze how the characteristics of participants in educational and social learning forums and the quality of events influence expectations and satisfaction of forums. The study also aims to provide strategic implications for forum organizers and give them suggestions on how to set up target audience, manage forum contents, speakers, and services, improve attendee satisfaction, and ultimately maximize overall outcomes. As exchanges among individuals, enterprises, and organizations, as well as countries are growing rapidly, the convention industry has become a key player in the market. Conventions have also become a venue for people to discuss a specific agenda or topic, exchange information and learn knowledge and insights. Especially, the forum - as part of the convention industry - plays a vital role providing educational and social learning opportunities as scholars and expertise come together to share their knowledge and experience through a variety of discussions. With its role, many of forums are taking place in recent years; however, there have been few empirical studies upon the forum itself. Also, there have been few attempts to research how the quality of forums affect participants' satisfaction along with their characteristics and how much of practical knowledge is provided throughout the events. This study is meaningful in that it is the first practical study that takes a deep understanding of the forum and sees how the quality of the forums influences participants' satisfaction and whether the characteristics of participants have a moderating effect in increasing the level of satisfaction. Forum organizers could also take a strategic approach as their major concerns are to increase the number of participants and raise degree of satisfaction by providing significant information. There are four key elements that determine success or failure of a social learning forum. The four elements are contents, speakers, services, and participants. Content plays an important role in providing rich information and knowledge for participants. Speakers are the main knowledge providers who contribute to the forum's social learning role. Also, the services provided by forum organizers such as simultaneous interpretation services, program brochures, lunch and refreshments, and the overall design of event hall can also influence the level of participants' satisfaction. Lastly, the participants and their characteristics are important since they are the ones who receive knowledge from the providers. The results of this study show that the quality of forum (content, speaker, and services) has a decisive effect on the participants' satisfaction and there are some differences in expectation among the participants in the forum. Also, some groups of participants were more likely to be stimulated by the quality of forum when determining their satisfaction. The study is modeled after MBN Y Forum 2016 and its participants' characteristics. The forum is one of the most representative social learning forums of South Korea and its audiences are mostly young people. It has analyzed how the participants' characteristics influence their satisfaction by grouping them into ${\Delta}participants$ who have invited for free and those who paid for the entrance fee, ${\Delta}first-time$ participants and returning participants, ${\Delta}voluntary$ and involuntary participants, ${\Delta}participants$ who registered through web and those who did through mobile, and ${\Delta}participants$ who registered during pre-sale opens and those who registered during general opens.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.9
• /
• pp.17-25
• /
• 2023

In this paper, we propose a model that can perform human pose estimation through a MobileViT-based model with fewer parameters and faster estimation. The based model demonstrates lightweight performance through a structure that combines features of convolutional neural networks with features of Vision Transformer. Transformer, which is a major mechanism in this study, has become more influential as its based models perform better than convolutional neural network-based models in the field of computer vision. Similarly, in the field of human pose estimation, Vision Transformer-based ViTPose maintains the best performance in all human pose estimation benchmarks such as COCO, OCHuman, and MPII. However, because Vision Transformer has a heavy model structure with a large number of parameters and requires a relatively large amount of computation, it costs users a lot to train the model. Accordingly, the based model overcame the insufficient Inductive Bias calculation problem, which requires a large amount of computation by Vision Transformer, with Local Representation through a convolutional neural network structure. Finally, the proposed model obtained a mean average precision of 0.694 on the MS COCO benchmark with 3.28 GFLOPs and 9.72 million parameters, which are 1/5 and 1/9 the number compared to ViTPose, respectively.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.5
• /
• pp.19-26
• /
• 2020

This paper proposes a framework for creating acrobatic scenes such as soccer ball lifting using various users' videos. The proposed method can generate a desired result within a few seconds using a general video of user recorded with a mobile phone. The framework of this paper is largely divided into three parts. The first is to analyze the posture by receiving the user's video. To do this, the user can calculate the pose of the user by analyzing the video using a deep learning technique, and track the movement of a selected body part. The second is to analyze the movement trajectory of the selected body part and calculate the location and time of hitting the object. Finally, the trajectory of the object is generated using the analyzed hitting information. Then, a natural object lifting scenes synchronized with the input user's video can be generated. Physical-based optimization was used to generate a realistic moving object. Using the method of this paper, we can produce various augmented reality applications.

• Proceedings of The KACE
• /
• 2017.08a
• /
• pp.201-204
• /
• 2017

빅 데이터의 시대가 도래하면서 이전보다 데이터로부터 유의미한 정보를 추출하는 것에 대한 연구가 활발하게 진행되고 있다. 딥러닝은 텍스트, 이미지, 동영상 등 다양한 데이터에 대한 학습을 가능하게 할 뿐만 아니라 높은 학습 정확도를 보임으로써 차세대 머선러닝 기술로 각광 받고 있다. 그러나 딥러닝은 일반적으로 학습해야하는 데이터가 많을 뿐만 아니라 학습에 요구되는 시간이 매우 길다. 또한 데이터의 전처리 수준과 학습 모델 튜닝에 의해 학습정확도가 크게 영향을 받기 때문에 활용이 어렵다. 딥러닝에서 학습에 요구되는 데이터의 양과 연산량이 많아지면서 분산 처리 프레임워크 기반 분산 학습을 통해 학습 정확도는 유지하면서 학습시간을 단축시키는 사례가 많아지고 있다. 본 연구에서는 범용 분산 처리 프레임워크인 아파치 스파크에서 데이터 병렬화 기반 분산 학습 모델을 활용하여 모바일 빅 데이터 분석을 위한 딥러닝을 구현한다. 딥러닝을 구현할 때 분산학습을 통해 학습 속도를 높이면서도 학습 정확도를 높이기 위한 모델 튜닝 방법을 연구한다. 또한 스파크의 분산 병렬처리 효율을 최대한 끌어올리기 위해 파티션 병렬 최적화 기법을 적용하여 딥러닝의 학습속도를 향상시킨다.

• The Journal of Bigdata
• /
• v.2 no.2
• /
• pp.9-16
• /
• 2017

Global Platform Solution Company (aka Amazon, Google, MS, IBM) who has cloud platform, are driving AI and Big Data service on their cloud platform. It will dramatically change Enterprise business value chain and infrastructures in Supply Chain Management, Enterprise Resource Planning in Customer relationship Management. Enterprise are focusing the channel with customers and Business Partners and also changing their infrastructures to platform by integrating data. It will be Digital Transformation for decision support. AI and Deep learning technology are rapidly combined to their data driven platform, which supports mobile, social and big data. The collaboration of platform service with business partner and the customer will generate new ecosystem market and it will be the new way of enterprise revolution as a part of the 4th industrial revolution.