Turkish Journal of Electrical Engineering and Computer Sciences, Mar 22, 2019
In this paper, we propose a new accuracy measurement model for the video stabilization method bas... more In this paper, we propose a new accuracy measurement model for the video stabilization method based on background motion that can accurately measure the performance of the video stabilization algorithm. Undesired residual motion present in the video can quantitatively be measured by the pixel by pixel background motion displacement between two consecutive background frames. First of all, foregrounds are removed from a stabilized video, and then we find the two-dimensional flow vectors for each pixel separately between two consecutive background frames. After that, we calculate a Euclidean distance between these two flow vectors for each pixel one by one, which is regarded as a displacement of each pixel. Then a total Euclidean distance of each frame is averaged to get a mean displacement for each pixel, which is called mean displacement error, and finally we calculate the average mean displacement error. Our experimental results show the effectiveness of our proposed method.
The immediate adoption of deep learning models into domain-specific tasks for edge intelligence-b... more The immediate adoption of deep learning models into domain-specific tasks for edge intelligence-based services still poses several challenges to overcome. The first is efficiently constructing the most suitable neural network architecture amongst the numerous types of available architectures. Once addressing this challenge, the second is understanding how to gather knowledge to build efficient neural network models from the user's devices (i.e., smartphone) without affecting the user's privacy. And the third critical issue is minimizing the gap between estimated and actual performance while building models. In this work, we propose a novel framework for deploying deep learning models for domain-specific tasks called MetaFed, which combines population-based meta-learning and federated learning to resolve the three challenges mentioned earlier. MetaFed autonomously constructs the potential domain-specific models with the help of population-based meta-learning by utilizing the ...
Downtime Minimization for Real-time AI Service on Intelligent Edge Nodes: Micro-Renewal Method
2021 IEEE 4th International Conference on Electronics and Communication Engineering (ICECE)
As the innovation of computing infrastructure evolves to edge computing via cloud computing, inte... more As the innovation of computing infrastructure evolves to edge computing via cloud computing, intelligent devices such as robots, drones, and autonomous vehicles, which are mobile edge nodes, also surged. Since the edge nodes have limited resources, artificial intelligence services are provided based on lightweight containers. In addition, as intelligent edge node users increase and the categories of users become vast, in order to provide artificial intelligence services according to the situations of all users, data on each situation is collected, and it is necessary to continuously update the learning model. However, if the service is being provided, downtime is inevitable for the updated model to be applied to the service. Therefore, in this paper, we propose a micro-renewal method that minimizes the interruption of the service provided to users in real time when the learning model in the service is updated.
Orchestration-Based Task Offloading for Mobile Edge Computing in Small-Cell Networks
Proceedings of International Joint Conference on Computational Intelligence, 2020
To execute computation-intensive applications and stringent latency-critical tasks at resource co... more To execute computation-intensive applications and stringent latency-critical tasks at resource constraints smart mobile devices, mobile edge computing (MEC) in small-cell networks is one of the leading thought, where mobile devices will offload their computation-intensive tasks to the adjacent small-cell network for faster processing. Currently, some research work has been done for combining mobile edge computing and small-cell networks together. Existing researches mostly concentrate on the user to small base station (SBS) offloading and improving the radio access performance using optimization, while the computing capability of SBS-MEC server is ignored. In order to acquire superior performance, an efficient orchestration-based task offloading for mobile edge computing in small-cell networks is proposed in this paper where edge orchestrator collects all the information from the neighboring small-cell SBS-MEC server to decide for forwarding the workloads from overloaded SBS-MEC to nearby SBS-MEC with a light workload. Simulation results affirm that orchestration-based task offloading scheme offers the best results not only by reducing the task failure but also with a smaller task completion time compared to other approaches in small-cell networks.
With rapidly increasing pervasive computing and ubiquitous services, a lot of data is being gener... more With rapidly increasing pervasive computing and ubiquitous services, a lot of data is being generated. With the advent of Internet of Things (IoT), heterogeneous devices and objects can become part of it and generate data with different frequencies. It is becoming very difficult to handle all that data, specially at the IoT end. Cloud computing plays a very vital role here. IoTs can be integrated with cloud computing, to form Cloud of Things (CoT). By this, resource constrained devices would be free from doing extensive tasks. For the resource management of CoT, pricing and billing is one of the important aspects. In this paper, we present the way of estimating prices and perform billing for different types of cloud service customers (CSCs), for different services, through a mathematical model. We implemented our model using Java and evaluated its performance through CloudSim simulation toolkit.
Performance Analysis of Data Parallelism Technique in Machine Learning for Human Activity Recognition Using LSTM
2019 IEEE International Conference on Cloud Computing Technology and Science (CloudCom), 2019
Human activity recognition (HAR), driven by large deep learning models, has received a lot of att... more Human activity recognition (HAR), driven by large deep learning models, has received a lot of attention in recent years due to its high applicability in diverse application domains, manipulate time-series data to speculate on activities. Meanwhile, the cloud term "as-a-service" has essentially revolutionized the information technology industry market over the last ten years. These two trends somehow are incorporating to inspire a new model for the assistive living application: HAR as a service in the cloud. However, with frequently updates deep learning frameworks in open source communities as well as various new hardware features release, which make a significant software management challenge for deep learning model developers. To address this problem, container techniques are widely employed to facilitate the deep learning software development cycle. In addition, models and the available datasets are being larger and more complicated, and so, an expanding amount of computing resources is desired so that these models are trained in a feasible amount of time. This requires an emerging distributed training approach, called data parallelism, to achieve low resource utilization and faster execution in training time. Therefore, in this paper, we apply the data parallelism to build an assistive living HAR application using LSTM model, deploying in containers within a Kubernetes cluster to enable the real-time recognition as well as prediction of changes in human activity patterns. We then systematically measure the influence of this technique on the performance of the HAR application. Firstly, we evaluate our system performance with regard to CPU and GPU when deployed in containers and host environment, then analyze the outcomes to verify the difference in terms of the model learning performance. Through the experiments, we figure out that data parallelism strategy is efficient for improving model learning performance. In addition, this technique helps to increase the scaling efficiency in our system.
BcN is a high-quality broadband network for multimedia services integrating telecommunication, br... more BcN is a high-quality broadband network for multimedia services integrating telecommunication, broadcasting, and Internet seamlessly at anywhere, anytime, and using any device. BcN is Particularly vulnerable to intrusion because it merges various traditional networks, wired, wireless and data networks. Because of this, one of the most important aspects in BcN is security in terms of reliability. So, in this paper, we suggest the sharing mechanism of security data among various service networks on the BcN. This distributed, hierarchical architecture enables BcN to be robust of attacks and failures, controls data traffic going in and out the backbone core through IP edge routers integrated with IDRS. Our proposed anomaly detection scheme on IDRS for BcN service also improves detection rate compared to the previous conventional approaches.
Energy efficiency oriented migration scheme in cloud data center
2017 IEEE International Conference on Cybernetics and Computational Intelligence (CyberneticsCom), 2017
Development of huge Cloud Data Centers (CDCs) and large-scale service application has led to enor... more Development of huge Cloud Data Centers (CDCs) and large-scale service application has led to enormous energy consumption. Migration objective used in many areas such as power reduction and load balancing in CDCs. However, most of the previous researches concentrated on maximize migration performance. Also, previous researches generally did not consider multi-objective. We have to consider combining multiple metrics and find good balance between metrics to minimize trade-off. In this paper, we proposed migration scheme for energy efficiency. The proposed scheme performs better than 9.5% the others.
Intelligent surveillance systems enable secured visibility features in the smart city era. One of... more Intelligent surveillance systems enable secured visibility features in the smart city era. One of the major models for pre-processing in intelligent surveillance systems is known as saliency detection, which provides facilities for multiple tasks such as object detection, object segmentation, video coding, image re-targeting, image-quality assessment, and image compression. Traditional models focus on improving detection accuracy at the cost of high complexity. However, these models are computationally expensive for real-world systems. To cope with this issue, we propose a fast-motion saliency method for surveillance systems under various background conditions. Our method is derived from streaming dynamic mode decomposition (s-DMD), which is a powerful tool in data science. First, DMD computes a set of modes in a streaming manner to derive spatial–temporal features, and a raw saliency map is generated from the sparse reconstruction process. Second, the final saliency map is refined ...
A medium access control (MAC) protocol is designed to disseminate safety messages reliably and ra... more A medium access control (MAC) protocol is designed to disseminate safety messages reliably and rapidly to improve the safety and efficiency of vehicles on the road in vehicular ad-hoc networks (VANETs). VANETs, which are created by moving vehicles, have specific properties, such as high node mobility with constrained movements and quick topology changes. Hence, MAC protocols should be designed to adapt to the changing data traffic patterns due to vehicle densities in the VANET environment. The latest multi-channel MAC protocols based on IEEE 802.11p and IEEE 1609.4 standards have higher performance than that of single-channel MAC protocols in every key performance indicator. Specifically, the multi-channel MAC protocols, which adapt themselves to different vehicular traffic densities, can guarantee a bounded transmission delay of real-time safety applications and an increased throughput for nonsafety applications. In this paper, we focus on the three following perspectives: First, the multi-channel MAC protocols are studied under saturated and non-saturated data traffic conditions; Second, we study the multidimensional Markov chains (up to three dimensions) used in the MAC protocols; and Third, the considered Markov models are compared with real-life application requirements to improve the existing analytical models and protocol designs. Finally, we summarize our findings and discuss the open issues concerning multi-channel MAC protocols for VANETs as a part of the Intelligent Transportation System.
A remote display QoE improvement scheme for interactive applications in low network bandwidth environment
Multimedia Tools and Applications, 2017
Screen transmission is an essential part of Desktop as a Service (DaaS) which directly influence ... more Screen transmission is an essential part of Desktop as a Service (DaaS) which directly influence the quality of experience (QoE). In this paper, we propose a novel QoE improvement scheme that dynamically controls the quality setting of the image compression before the screen transmission to decrease response time of the system still maintaining the satisfactory image quality, hence improves the QoE in interactive applications in a band-limited environment. The proposed scheme first selects the best quality setting appropriate for current network bandwidth quota, then uses the remaining bandwidth to improve the quality setting of low motion regions without any adverse effect on response time. To enable the adaptive quality selection and image quality refinement, we propose a compressed image file size inference model and a block priority calculation method respectively. Particularly, we implement our QoE Improvement Scheme to work with screen content coding. Both quantitative measurements and users’ evaluations in the experiments show that our QoE improvement scheme improves QoS as well as QoE by utilizing the available network bandwidth efficiently.
Energy Management for Mobile Virtual Desktop Infrastructure in Mobile Cloud Environment
정보과학회논문지 컴퓨팅의 실제 및 레터, Aug 1, 2014
최근 모바일 가상 데스크톱 인프라(mobile virtual desktop infrastructure)이 모바일 장치에서 클라우드 자원을 활용하기 위한 해결 방법으로 등장하였... more 최근 모바일 가상 데스크톱 인프라(mobile virtual desktop infrastructure)이 모바일 장치에서 클라우드 자원을 활용하기 위한 해결 방법으로 등장하였다. 가상 머신을 원격으로 조정함으로써 컴퓨팅 작업이 모바일 장치에서 처리되는 대신 데이터 센터로 옮겨가게 되었다. 그러나 모바일 장치로 원격 데스크톱 서비스를 이용하는 것은 PC에서와는 달리 전화 통화, 메시지, 배터리, wifi/3G, 운영체제, 오류 등의 이유로 잠시 접속이 끊어질 수 있다. 에너지 소비 절약을 위하여 이와 같이 모바일 장치가 다시 서비스를 제공하기 전까지 동작하고 있는 가상 머신들은 절전 상태로 변경될 필요가 있다. 그리하여 본 논문에서 우리는 데이터 센터의 가상 머신 상태를 관리하기 위한 에너지 정책을 제안하였다. 성능 평가를 위한 시뮬레이션을 통해서 제안하는 관리 방법이 모바일 클라우드 환경에서 약 74.5%의 에너지 절약을 할 수 있음을 보였다.
A rule-based data grouping method for personalized log analysis system in big data computing
Fourth edition of the International Conference on the Innovative Computing Technology (INTECH 2014), 2014
Nowadays, providing personalized service to customers is one of the main issues in big data servi... more Nowadays, providing personalized service to customers is one of the main issues in big data services. To provide the personalized service, analyzing various logs and cooperation between data analysts and developers are critical. However, the problem is that overhead can occur when the log data is analyzed due to general characteristics of big data system as well-known 4Vs(Velocity, Various, Value and Volume). Also, generally it is hard for data analysts and developers to work together because they use different interfaces. Therefore, we propose a personalized log analysis system including rule-based data grouping method in order for the improved performance of personalized log analysis and more flexible cooperation between data analysts and developers. The evaluation of the proposed system performs well for cooperation and grouping along with the R SW tool.
HiLiCLoud: High performance and lightweight mobile cloud infrastructure for monitor and benchmark services
2015 17th Asia-Pacific Network Operations and Management Symposium (APNOMS), 2015
In the area of cloud infrastructure environment, the management tool to monitor and control the c... more In the area of cloud infrastructure environment, the management tool to monitor and control the cloud resources is the important factor that can drive the cost benefit of the cloud vendors. But most these tools are bundled within the high cost commercial platforms and are optimized to run on desktop computers. With the vision that Mobile Cloud Computing will be the future technology paradigm that dominates the IT industry, we want to create a cloud management tool that is open source, fast, lightweight and mobile friendly. We take the initial steps by implementing our framework using several popular technologies such as RESTful, Java Message Service, JSON, and we call it “High performance and Lightweight Mobile Cloud Infrastructure Monitor and Benchmark Service” or HiLiCloud. The initial testings show competitive evaluation results.
M2M Emergency Help Alert Mobile Cloud Architecture
2015 IEEE 29th International Conference on Advanced Information Networking and Applications Workshops, 2015
Emergency situations are unfortunately part of our lives. Today's smart computing allow us ha... more Emergency situations are unfortunately part of our lives. Today's smart computing allow us handle such situations and fulfill our requirements more efficiently and effectively. This paper presents architecture to handle various kinds of emergency situations more efficiently and effectively, by allowing the user (victim or witness) easy and quick way to alert the concerned department (s) with just a single button press. The service automatically sends the location of incident and contacts the appropriate emergency dealing department automatically through already stored contact numbers. The emergency related information is then synchronized automatically to the mobile cloud, allowing further analysis and improvement in safety of people and creates further services for the concerned authorities and users. Performance in most certain scenarios is also evaluated and presented in this study.
In recent years, Knowledge Distillation has obtained a significant interest in mobile, edge, and ... more In recent years, Knowledge Distillation has obtained a significant interest in mobile, edge, and IoT devices due to its ability to transfer knowledge from the large and complex teacher to the lightweight student network. Intuitively, Knowledge Distillation refers to forcing the student to mimic the teacher's neuron responses to improve the generalization of the student by deploying the distillation losses as the regularization terms. However, the non-linearity of the hidden layers and the high dimensionality of the feature maps make the knowledge transfer a rigorous task. Though numerous methods have been proposed to transfer the teacher's neuron responses in the form of diverse feature characteristics such as attention, contrastive representation, and so on, to the best of our knowledge, no prior works considered feature-level non-linearity during distillation. In this work, we ask, does feature-level non-linearity-based approaches can improve student performance? For investigating those concerns, we propose a novel knowledge distillation technique called the NeuRes (Neuron's Responses) via distilling the Sparse Activation Maps (SAMs) to transfer the highly activated Neurons Responses to the student to enhance the representation capability. Proposed NeuRes selects the highly activated neuron responses that produce Sparse Activation Maps (SAMs) while transferring the knowledge based on activation normalization. Our proposed NeuRes also transfers the translation invariant features using auxiliary classifiers and augmented data to improve students' generalization. The detailed ablation studies and extensive experiments on model compression, transferability, adversarial robustness, and few-shot learning verify that NeuRes outperforms state-of-the-art distillation techniques on the standard benchmark datasets.
Uploads
Papers by Eui-nam Huh