The long-term use of machine learning models can result in degraded performance due to data drift and other factors. We have previously proposed a data selection mechanism for time-series data of machine learning models. When data drift occurs, the models have to learn again using large-scale stream data. Thus, it is important for machine learning algorithms to introduce a mechanism avoiding useless data. This study examines the effect of data selection with an adversarial classifier using synthetic data.

The CPU is one of the most battery-consuming elements in a smartphone. The power consumption and user experience are in a trade-off relationship. These can be controlled by the CPU clock rate. Increasing the clock rate makes the user experience better and power consumption worse. On the contrary, the user experience declined and power consumption savings is improved by decreasing the rate. Thus, the clock rate should be increased only when large CPU resources are required. In our previous work, we proposed a concept of increasing the clock rate at an invocation of a CPU-consuming and time-consuming method. However, we did not discuss whether there are method invocations that meet these conditions. In this paper, we investigate the CPU usage trend of practical Android applications and show that many applications have method invocations suitable for the proposed dynamic CPU clock rate adjustment.

This paper provides an overview of tourist information distribution system that sends information corresponding to places and times. We completed the system successfully, although it is difficult to clearly extract information of the date, time, place, and event name from non-structured data written in natural language such as the language used on the SNS. We evaluated how many pieces of information are collected.

SNS websites and messages have become a regular part of daily life. However, some large-scale disasters such as earthquakes could damage network cables and equipment and cause network failures. Despite the assistance of information related to disasters and support that is effective for controlling the situation and offering help, it is possible that frequently used SNS messages and websites will become unreachable in a disaster area. Moreover, in a disaster area, the network could break into several sub-networks because of the damage of the backbone network. It is possible that the sub-networks can provide network services in a smaller area range without connecting to the Internet. Therefore, this paper proposes a decentralized SNS system that could solve two major problems during large-scale disasters. First, it describes a stable and continuous SNS network for the affected population and helpers without the influence of networks in the disaster region. Second, it proposes a reliable method to share synchronized data across the affected areas. In this case, a reliable method to control connections among shelters, whether newly built shelters or shelters that lost connection during the disaster, should also be considered. This paper shows the effectiveness of the proposed decentralized SNS system using XMPP to provide SNS for refugees and helpers in each shelter under an emergency of network failure. The connection control among shelters has also been examined and was designed to provide secure connections and data synchronization for this decentralized SNS system.

The spread of various sensors and the development of cloud computing technologies enable the accumulation and use of many live logs in ordinary homes. In addition, deep learning technologies have been widely used for image and speech recognition processing. However, a key issue for deep learning is heavy processing loads. To operate a service that utilizes sensor data, those data are transmitted from sensors in ordinary homes to a cloud and analyzed in the cloud. However, services that involve moving image analysis require large amounts of data to be transferred continuously and high computing power for the analysis; hence, it is difficult to process them in real time in the cloud using a conventional stream data processing framework. First, we perform preliminary experiments using Apache Spark [3] (hereinafter called Spark), which is a representative cluster computing platform that is designed to be fast and versatile, and Ray [4] , which is a distributed execution framework. We investigate the characteristics of their distributed recognition processing and demonstrate that Ray enables scalable distributed processing. Next, We implement a prototype system of the proposed distributed stream processing infrastructure using Ray and Apache Kafka [1] (hereinafter called Kafka), which is a distributed messaging system, and demonstrate its performance.

As the use of various sensors and cloud computing technologies has spread, many life-log analysis applications for safety services for the elderly and children have been developed. However, it is difficult to perform real-time large data processing in clouds due to the computational complexity of the analysis because efficient deployment schemes of streaming computing components over cloud resources have not been well-investigated. In this study, we propose a video analysis framework that collects videos from multiple cameras and analyzes them using Apache Kafka and Apache Spark Streaming. We first investigate the data transfer performance of Apache Kafka and examine efficient cluster configuration and parameter settings. We then apply this configuration to the proposed framework and measure the data analysis throughput. The experimental results show that the overall throughput varies depending on the number of broker nodes that store data, the number of topic partitions of data, and the number of nodes that conduct analysis processing. In addition, it is confirmed that the number of cores is needed to consider for the efficient cluster configuration, and that the network bandwidth between the nodes becomes a bottleneck as the amount of data and the number of components increase.

The Spread of cameras and sensors and cloud technologies enable us to obtain life logs at ordinary homes and transmit the captured data to a cloud for life log analysis. However, the amount of processing for video data analysis in a cloud drastically increases when a very large number of homes send data to the cloud. In this research, we aim to improve the efficiency of distributed video data analysis processing by using the parallel deep learning framework Chainer [2] and the distribution processing platform Apache Spark [1] (Spark). In this paper, we construct a Spark cluster and investigate the performance of parallel data processing using Spark varying parameter settings.

When the Great East Japan Earthquake occurred in 2011, the network connectivity was significantly degraded in the wide area due to the multiple network failures as well as the traffic congestion. When the network failures occurred in multiple areas, it was difficult to quickly recognize the entire network situation only using the network traffic monitor system. In our prior works, we found that SNS messages contain the useful information to recognize the big picture of the network failures and proposed the network control system using SNS messages to improve the quickness of the network recovery. As the another critical issue in case of a large-scale disaster, users could not obtain the emergency information due to the network disturbance because the current IP network is operated not being aware of the applications. Thus we propose the application specific traffic control system with failure detection function based on SNS message to prioritize the important application traffic in the event of the large-scale disaster. Based on a series of experiments, this paper shows the effectiveness of a system that detects connection failure based on social information and controls the network bandwidth for each application. Especially, we focus on application specific traffic control experiment. An automatic SDN control is performed with the network virtualization node FLARE having SDN extension capability as well as the network slicing capability. We perform the experiments to determine the type of application based on the traffic and perform bandwidth control for each application using real Internet applications.

When the Great East Japan Earthquake occurred in 2011, the network connectivity was significantly degraded in the wide area due to the multiple network failures as well as the traffic congestion. When the network failures occurred in multiple areas, it was difficult to quickly recognize the entire network situation only using the network traffic monitor system. In our prior works, we found that SNS messages contain the useful information to recognize the big picture of the network failures and proposed the network control system using SNS messages to improve the quickness of the network recovery. As the another critical issue in case of a large-scale disaster, people could not obtain the emergency information due to the network disturbance because the current IP network is operated not being aware of the applications. Thus we propose the specific application traffic control system with failure detection function based on SNS message to prioritize the important application traffic in the event of the large-scale disasters. Based on a series of experiments, this paper shows the effectiveness of a system that detects connection failure based on social information and controls the network bandwidth for each application. Especially, we focus on application specific traffic control experiment. An automatic SDN control is performed with the network virtualization node FLARE having SDN extension capability as well as the network slicing capability. We perform the experiments to determine the type of application based on the traffic and perform bandwidth control for each application using some real Internet applications.

The spread of various sensors and Cloud technologies has made it easy to acquire life-logs and accumulate data. As a result, many life-log analysis applications, which transfer data from sensors, especially cameras to a Cloud and analyze them in the Cloud, have been developed. Cameras with a server function called network cameras have become cheap and readily available for security services and the monitoring of pets and children from remote locations. In these services, raw data from sensors, including cameras, are generally transferred to a Cloud and processed there. However, it is difficult to transfer raw data from sensors to a Cloud because of the limitation of network bandwidth between sensors and a Cloud and privacy issues caused by sending raw sensor data to a Cloud.

Personal computer/workstation (PC/WS) clusters have come to be studied intensively in the field of parallel and distributed computing. From the viewpoint of applications, data intensive applications including data mining and ad-hoc query processing in databases are considered very important for massively parallel processors, in addition to the conventional scientific calculation. Thus, investigating the feasibility of such applications on a PC cluster is meaningful. A PC cluster connected with a storage area network (SAN) is built and evaluated with a data mining application. In the case of a SAN-connected cluster, each node can access all shared disks directly without using a LAN; thus, SAN-connected clusters achieve much better performance than LAN-connected clusters for disk-to-disk copy operations. However, if a lot of nodes access the same shared disk simultaneously, application performance degrades due to the I/O-bottleneck. A runtime data declustering method, in which data is declustered to several other disks dynamically during the execution of the application, is proposed to resolve this problem.