論文

Recently, the development of data communication networks and advances in computer processing capacity have significantly increased the amount of data to be managed. Although future innovations using data are expected, privacy violation of data has become a problem. For example, image disclosure in Social Networking Service may result in infringement of portrait rights. Previously, studies on anonymization have been conducted to disclose data for protecting personal information. However, conventional anonymization, even for high-dimensional data such as faced images, uses an averaging operation in anonymization and does not consider complex relationships between dimensions. Therefore, the loss of semantic meaning increases because it is caused by the anonymization process, assuming an Euclidean data space. We herein propose Multidimensional Inputs K-anonymizing Unit(MIKU), an anonymization algorithm focusing on face image anonymization as a typical example of high-dimensional datMIKU enables anonymization that retains the image quality. As the comparison method directly anonymizes images, the quality of images is deteriorated. MIKU considers the relation between dimensions using the latent space of StyleGAN. The effect of using latent space was confirmed by comparing the quantitative results of Fréchet Inception Distance(FID) and the qualitative evaluation of the output image with the comparison method. The quantitative results by FID proved that MIKU achieved better performance when K-anonymity was large. When K-anonymity is small, we clarified that the image quality of the comparison method such as FID could not be measured correctly and was overestimated because of the effect of the inception model's linearity. In addition, the qualitative evaluation of anonymized images shows that MIKU generates a more natural image in the contour and hair expressions than the comparison method, and the anonymized images of MIKU contain no unnatural edge lines on a face such as those generated in the comparison method.

Based on the development of networking services and the emergence of various computing environments, such as edge and cloud environments, appropriate service allocation based on processing delays, computational resources, data volumes, and other requirements has become crucial. To support dynamic load shifting and reduce service downtime, service allocation should be performed using live migration, which maintains network consistency. There have been many studies on live migration techniques and their implementation in software applications. However, limited research has been reported on field-programmable gate array (FPGA)-based transparent service migration. This paper proposes and evaluates a transparent live-migration architecture for an FPGA computing system that maintains network consistency before and after migration and enables effective migration without any software or hardware extensions on communicating edges or cloud nodes. The proposed architecture achieves sufficient throughput and sufficiently low downtime to be introduced into cloud and edge nodes.

Named Data Networking (NDN) is a promising inter-networking paradigm that focus on content rather than hosts and their physical locations. In NDN Consumers issue Interests for Contents. Producers generate a content in response to each received interest and such content is routed back to the requesting consumer. When compared to IP, NDN brings advantages such as better throughput and lower latency, because routers are able to cache popular contents and satisfy interests for such contents locally. However, before being considered a viable approach, NDN should offer security services that are ideally better, but at least equivalent to current mechanisms in IP.In this regard, mechanisms to prevent DDoS are of paramount importance. In this work we propose FROG: a simple yet effective Interest Flooding Attack (IFA) detection and mitigation method. FROG runs on routers that are directly connected to NDN consumers and monitors packet hop counts. It then calculates mean and variance using stored hop counts to distinguish attackers from legitimate users. We use the NDN simulator ndnSIM to evaluate FROG's effectiveness. Our results show that FROG improves resilience against DDoS attacks. In particular, during an attack, legitimate users can still receive 75% of requested contents. Without FROG this number decreases to 50%.

With the increase in the number of Internet of Things (IoT) devices, many applications using the data accumulated from IoT devices have been proposed. Consequently, privacy violation due to data disclosure is becoming a problem. Data anonymization is a technique for publishing data while preserving data privacy and is an effective method for solving the problem of privacy violation. There are several requirements for implementing anonymization systems suitable for IoT devices, where the amount of hardware and software resources, function extensibility, and processing throughput are limited. To satisfy these requirements, anonymization hardware for anonymizing data packets from IoT devices is proposed. The proposed hardware is allocated on the intermediate location of network and anonymizes the data packet stream from IoT devices transparently. The proposed anonymization process does not affect data communication or routing. This transparency allows an additional anonymization function to be installed on all types of IoT devices without modifying the devices. The proposed hardware allows IoT devices to anonymize the data packet stream. To reduce the latency and achieve transparent anonymization at a low cost, the proposed anonymization hardware was implemented using a field-programmable gate array. The throughput and latency of the testbed of the proposed hardware were 1,286 Mbps and 330 ns, respectively. The power consumption of the proposed hardware was lower than that for a software implementation.

To prevent non-communicable diseases, it is important to review consumers' dietary habits. Most existing applications for improving eating habits require users to upload photos of their meals and record them manually, but such procedures are time-consuming and laborious. Because the targets of the proposed method are those who are not highly conscious of their health, it is necessary to make the application easy to use. This study applies the history of supermarket purchases to calculate nutrient intake and recommend foods that improve nutritional balance with the least amount of user input. Consequently, the nutrient intake can be estimated with an acceptable error, and foods that are easy for the user to purchase can be recommended. Because the proposed method does not use artificial intelligence technologies to generate recommendations, the reasons for food recommendations are clear and the computational cost is reduced.

In recent years, a large number of Internet of Things (IoT) devices have appeared. Accordingly, various services using data from such devices have been proposed. However, the collected raw data include private information, and thus, privacy problems arise. Data anonymization is a method for removing privacy-sensitive information from raw data. Data anonymization for IoT data services should satisfy the following requirements. First, the raw data should be anonymized between a device and the cloud server. Second, the anonymization methods and the destinations of the collected data should be flexibly configured, as they depend on data types and agreements with data suppliers. Third, network transparency is necessary for ease of installation. However, conventional data anonymization systems do not satisfy these requirements. We propose anonymization hardware that functions as a network router on network edges. It directly anonymizes data in network packets. Moreover, it decides the destination IP address of the packets and anonymizes data based on their content. For high-Throughput and low-power processing of the packets, the proposed hardware was implemented by using a field-programmable gate array. The throughput of the proposed hardware achieved 10 Gbps wire speed, and the power consumption was lower than that of software implementation.

The rapid growth in cloud computing, the Internet of Things (IoT), and data processing via Machine Learning (ML), have greatly increased our need for computing resources. Given this rapid growth, it is expected that data centers will consume more and more of our global energy supply. Improving their energy efficiency is therefore crucial. One of the biggest sources of energy consumption is the energy required to cool the data centers, and ensure that the servers stay within their intended operating temperature range. Indeed, about 40% of a data center’s total power consumption is for air conditioning[1]. Here, we study how the server air inlet and outlet, as well as the CPU, temperatures depend upon server loads typical of real Internet Protocol (IP) traces. The trace data used here are from Google clusters and include the times, job and task ID, as well as the number and usage of CPU cores. The resulting IT loads are distributed using standard load-balancing methods such as Round Robin (RR) and the CPU utilization method. Experiments are conducted in the Data Center Laboratory (DCL) at the Georgia Institute of Technology to monitor the server outlet air temperature, as well as real-time CPU temperatures for servers at different heights within the rack. Server temperatures were measured by on-line temperature monitoring with Xbee, Raspberry PI, Arduino, and hot-wire anemometers. Given that the temperature response varies with server position, in part due to spatial variations in the cooling airflow over the rack inlet and the server fan speeds, a new load-balancing approach that accounts for spatially varying temperature response within a rack is tested and validated in this paper.

Road traffic is a supporting medium that is essential to various means of land transportation that connect various locations and even large cities. Smooth traffic is a good parameter of the management of a city, also as a sign to determine the technological and economic development in the city. As the rate of civilization grows, as well as the number of people, the traffic becomes more crowded, so that congestion occurs in to an alarming level. The solution to solve the problem is by applying some methods as well as traffic engineering. However, implementing such methods without proper planning will lead to higher costs and lesser optimal results. In this article, we proposed a traffic simulation system called as 'LINTAS', a simulator tool in an attempt to resolve the traffic jam. The exact method and traffic engineering, before it is tested in a real situation, is first simulated through the LINTAS system, so that it can be known its effectiveness. LINTAS are built using the SimEvents toolbox and runs on MATLAB-Simulink software. The LINTAS simulator is performed based on the mathematical sciences, especially the field of queueing theory.

Indoor occupancy estimation is a critical analytical task for several applications (e.g., social isolation of elderlies). The proliferation of Internet of Things (IoT) devices enabled the occupancy estimation, as it provided access to a mass amount of data. Several works have been proposed exploiting the IoT Passive Inference (PIR) or environmental (e.g., CO2) features. These works however are traditionally selecting the feature space at the learning phase and passively using it over time. Hence, they ignore the dynamics of indoor occupancy, such as the location of the occupant or his motion patterns, leading to a decreasing accuracy over time. In this paper, we study those dynamics and show that motion patterns, along with environmental features favor the occupancy estimation. We design a Location-Aware Hidden Markov Model (HMM), which dynamically adapts the feature space based on the occupant's location. Our experiments on real data show that Location-Aware HMM can reach up to 10% better accuracy than Conventional HMM.

Smart communities, smart cities, smart towns, and similar smart terminologies are trending recently, along with the Internet of Things (IoT) and the Cloud, as popular new information and communication technologies (ICTs). This paper proposes a synergistic solution to urbanization problems from a technological perspective that is beneficial both to residents and to local government. Case studies of a smart community implementation in Saitama city are also provided, from the viewpoint of integrated smart infrastructure services.

The increasing demand for cloud computing services to more flexibly use computing resources has led to a severe problem of excessive energy consumption in data centers. In recent research, the energy consumption of computing and cooling equipment has attracted much attention. To reduce the energy consumption in data centers, the energy efficiency of the cooling equipment needs to be considered. This study theoretically modeled the relations between the server's internal heat sources and exhaust temperatures. First, we proposed a model that expands the previous study to describe the temperature change delay and to apply it to multiple internal heat sources. Then, we experimentally evaluated the prediction accuracy of the proposed model using a real server and conducted a simulation to confirm the efficiency of this prediction model for load balancing. The simulation results showed that the proposed method reduced the average maximum server temperature by 0.27 °C compared to the conventional method, which theoretically leads to energy savings of 260 kW per day and 95 MW per year in typical data centers.

Preserving an individual's privacy when publishing data are essential, and anonymization has been getting attention as the solution. When anonymizing data, it is necessary to contemplate the possibilities of linkage with other data which can lead to privacy violation. Trajectory data are one of the data, which contains personal data. Consequently, various anonymization methods of trajectory data have been considered by researchers. However, most research handle trajectory data as polylines connecting location data or as a sequence of location data. In other words, it lacks on considering the connection with map data. In this paper, we will consider the anonymization of trajectory data of moving users matched according to map data, which we will be calling pathing data. According to k-anonymity principle, data can be published if there are k of the same data. We will use k-anonymity principle to quantitively judge the risk of privacy violation and propose two methods that can fulfill the anonymization requirements with low data loss. The two methods are Map Matching to Node (MMtoN) and Map matching to Edge (MMtoE), which judges k-anonymity by segments of pathing data.

There has been rapid growth in the number of Internet of Things (IoT) devices that produce a large volume of data such as location data, temperature data, and power usage data. These data are used in various types of services on IoT platforms. IoT services such as autonomous vehicles and factory automation system have the following three requirements. First, confidential data such as machine ID, power usage, and location information need to be locally anonymized before they are sent to a cloud server. Second, these applications require low latency response of less than 10 ms. Third, the bandwidth usage that an enormous number of IoT devices generate needs to be efficiently reduced to alleviate the burden on the cloud server. However, existing IoT platforms have limitations on the configuration of applications because most of them are cloud-based. In this study, we propose a fog-based IoT platform, where fog nodes achieve network transparency for the IoT devices and the cloud server. The network-transparent machine can be installed without modification of the existing network configuration and routing. Fog nodes transparently achieve the three requirements: anonymization of specific data in packets, real-time feedback, and reduction in the bandwidth usage.

Software defined networking (SDN) and network function virtualization (NFV) are proposed as software based applications to cater to smart services requirements of smart communities. The services are linked together to support different sets of clients using service function chaining (SFC). Service functions (SFs) in an SFC must be distributed among available computing resources by creating a service function path (SFP) allowing resource management and optimal SF execution. This paper proposes a novel SFP allocation algorithm considering the computation capabilities of hardware resources while minimizing the completion time of SFCs. The proposed algorithm shows 10% performance increment compared to recently developed algorithms such as nearly optimal service function path and optimal service function selection algorithms.

In heating, ventilation, and air conditioning control, it is crucial to maintain the comfort of residents. Thermal comfort is typically assessed using the predicted mean vote (PMV) index. PMV depends on six factors: air temperature, mean radiant temperature, air velocity, air humidity, metabolic rate, and clothing insulation. Although PMV can be estimated by measuring these factors directly, this process is costly because multiple sensors are required. Furthermore, measuring metabolic rate and clothing insulation is especially costly because expensive and complex sensors are required. To solve these problems, this paper proposes a practical method for estimating PMV by estimating metabolic rate and clothing insulation using a low-cost infrared array (IrA) sensor. In this study, an IrA sensor called 'Grid-EYE' is adopted. PMV parameters other than air velocity and humidity can be measured when the proposed method and an IrA sensor are implemented in an air conditioner. Human detection is done using the temperature map captured by the sensor and their PMV values are estimated individually. Heat sources around people are also detected and their influence on PMV estimation is evaluated. Practical experiments demonstrate the validity of the proposed method by providing estimated PMV values close to theoretical values and real sensations. Therefore, the proposed method can contribute to providing comfortable living spaces and improving energy consumption and amenities efficiently.

Recently, the number of Internet of Things (IoT) sensors has been increasing rapidly; hence, various data are gathered. As a secondary use of the data, they are useful in providing new services, such as the demand response service in the Smart Grid. However, data services cause several problems in preserving privacy and during computation. This study focuses on these two significant problems. First, the invasion of privacy while using the data to provide such new services is problematic. A lot of private information is available in the data. For example, power consumption data may reveal the lifestyle of the residents, and the technique of obtaining information is known as nonintrusive load monitoring. Second, the penetration of IoT devices and sensors increases the computational and communicating energy consumption for processing the data and for providing various services using the data. In this paper, a new method is proposed to solve these two problems. This method is based on the fact that the anonymization process reduces the amount of information itself, as well as the quantity of computational resources required. This leads to a trade-off between anonymization level and computational cost. For example, raw data have a maximum amount of information and maximum computational cost. In contrast, fully generalized data (all zero data) have minimum amount of information and minimum computational cost. Compared to the conventional method, the proposed method demonstrated lower precision and a higher error rate. Therefore, the proposed method aims to control the trade-off and enables the provision of anonymized data with less information, the required anonymity level, and low computational cost compared to the conventional method. The proposed method is practiced using power consumption data gathered from the Urban Design Center Misiono (UDCMi) and the demand response service is evaluated as an experiment using the data. In this evaluation, a simple model of energy consumption was used in the calculation, which uses the required bit width of the arithmetic logic unit (ALU) for providing the service. The computational efficiency of the proposed method was increased by 60% when k = 2 and by 65% when k = 3, 4, 5, 6 compared to the conventional method. The method can also maintain an acceptable range of service error. The transprecision platform can restrict energy consumption by reducing the bit width of the data. Therefore, the proposed anonymization method can also reduce energy consumption by achieving lower usage of the ALU based on the transprecision architecture.

The development of IoT has led to the creation of a data-enriched environment that enables data gathering by using distributed sensors and terminals. However, in this environment, the cost of data analysis has increased. Machine learning has gained attention for reducing the cost because enabling automatic data analysis, as well as multidimensional data, is expected. However, for enormous data, such as Big Data, we still have to pay costs. Therefore, selecting feature values when using machine learning technology is essential, especially as inputs of a classifier. Selecting the feature values increases its estimation accuracy. Moreover, the time cost, as well as calculation cost, needs consideration for the actual time-critical use of machine learning, especially in its learning process. Therefore, in this study, we proposed an algorithm that selected suitable feature values in required time. The proposed method consists of two stages: stepwise input selection stage using ANOVA and feature deletion stage according to the contribution rate of the features to estimate accuracy. These selection and deletion processes continue until the required processing time. We confirmed the efficiency of the proposed method by using an environment of a crystallization process in a factory and a household's occupancy estimation. A comparison with the original stepwise input method proved that the proposed method improved the estimation accuracy by 2% and 5% in the estimation of the substance amount of the crystallization process and household's occupancy, respectively.

Home energy management system (HEMS), enabled by the development of the Internet of Things (IoT), issue behavior change recommendations to encourage residents to reduce their energy consumption. Receiving these suggestions from HEMS makes it easier for them to set specific reduction goals and raise their awareness of energy saving. This feedback will lead to effective power reduction in the household sector. However, each user has unique preferences, and uniformly generated recommendations may not be followed if they do not match the preferences of the specific user. In addition, frequent recommendations that are not aligned with their preferences may stress users and decrease their motivation to reduce energy consumption. This paper presents a practical method of making behavior change recommendations reflecting users' response rates and considering their stress. Targeting the action of opening a window, we illustrate how our system induces behavioral change. To increase the users' response rate and reduce their stress, we adjust the recommendation for each user from two perspectives. First, assuming that users open windows mainly depending on the external temperature, humidity, wind, and weather, we introduce the k-nearest neighbors (k-NN) classification using these parameters as the explanatory variables to predict the possibility that the user accepts the window-opening recommendation. Generating recommendations only when the predicted probability is high enables building a unique recommendation system considering user preferences. Second, if the recommendations are sent frequently, users may become tired of following them; this leads to a situation in which users ignore recommendations or turn off their notifications. To avoid such a situation, we propose adjusting the delivery interval according to the users' response rate. When we schedule the notification cycle, we introduce a forgetting curve, assuming that the users' stress on the recommendation decreases over time. We conducted a simulation using historical weather data. The response rate and thermal sensation of users with different variations were set, and the delivery timing of the recommendation was changed according to these factors. The proposed methods are expected to effectively generate behavioral changes by having users take medium-to long-term initiatives without lowering their motivation.

The agricultural population in Japan is aging and shrinking. Therefore, it is necessary to improve the efficiency of farm work. Additionally, it takes many years to learn the skills required for farming, which discourages many young people from joining the field. As a countermeasure to this problem, smart agriculture technology is gaining increasing attention. Smart agriculture is an approach to increasing the quantity and quality of crops by leveraging advanced technologies. In smart agriculture, a large amount of data are collected from various sensors and internet of things devices. As the use of information and communication technology in agriculture spreads, the demand for data sharing between different agricultural systems increases accordingly. However, electronic data exchange interfaces for agricultural data are not standardized. Therefore, in Japan, it has been recommended to unify data formats using extensible markup language (XML). In this study, we focused on data regarding crop growth states and developed a method for representing plant structure using XML. In the proposed method, branches and fruits are extracted from plant images and their connections are expressed using the hierarchical structure of XML. Compared to conventional management of crop growth states based on images, the proposed method significantly reduces data size. Furthermore, because XML elements can be easily searched and sorted using XPath and XQuery, the XML format makes data easy to utilize for many services. For example, by counting numbers of fruits, profits and work times can be predicted. Additionally, data regarding plant structure is useful for directing farmers or robots to harvest fruits. The proposed method contributes to improving productivity and helps inexperienced farmers.

Recently, data traffic on the Internet has increased due to the rapid growth of various Internet-based services. The convergence of user requests means that servers are overloaded. To solve this problem, service providers generally install multiple servers and distribute requests using a load balancer. The existing load balancing algorithms do not estimate the size of the load of unknown requests. However, the requested contents are heterogeneous and complex, so the size of the load is dependent on the servers and the contents of the requests. In this study, we propose a load balancing algorithm that distributes the requests based on estimates of the processing time, which avoids mismatches between the characteristics of servers and the request contents. The processing time for requests is estimated based on the requested contents by online machine learning, and a strategy to cover the latency of machine learning is proposed and partially conducted. To test the algorithm, we built a model of multiple database servers and performed an experiment using real log data for database requests. The simulation results showed that the proposed algorithm reduced the average processing time for requests by 94.5% compared with round robin and by 28.3% compared with least connections.