Research into mimicry of the functioning and behavior of living creatures ("biomimetics") has resulted in many inventions, such as hook-and-loop fasteners (inspired by cocklebur) and soccer goal nets (inspired by the shape of honeycombs). In the field of information and communication technology, too, many studies have applied the behavior of living creatures to problems such as networking. Recently, as networks have become larger and more complicated, it has become more difficult to understand the global status of networks and to control them appropriately. Toward solving this type of problem, we focus on swarm intelligence, which is seen in flocking birds and schooling fish. This swarm intelligence is social behavior arising from the autonomous behaviors of individuals, and we believe that applying this concept will enable distributed and adaptive network control.
We focus on communication between living creatures. Specifically, we plan to apply mathematical models to acoustic communication among Japanese tree frogs in order to model a network. To do so, we estimate the location of frogs and conduct experiments in real environments where frogs live, such as the Oki Islands of Japan.
An additional focus of our laboratory is understanding suitable mappings between groups in the natural world and networks in the Internet of Things (IoT). IoT networks include many and varied sensors and devices, and control of such networks could benefit from the advantages of biological systems, which achieve scalability, robustness, and adaptability under conditions of environmental change. We are particularly interested in models of the collective decision-making of groups with leaders, which may allow network control that can reach target states more quickly by designating leader nodes to guide the behavior of other nodes.
Human brains are very good at solving problems. It is reported that it takes the K computer (a massive supercomputer cluster) 40 minutes to reproduce a single second of human brain activity. The brain carries out extremely complicated processing but uses only a fraction of the energy used by a supercomputer to do so. The human brain is composed primarily of connections among neurons, and the resulting network has high communication efficiency and robustness. We believe that these characteristics of the brain network can be mimicked to construct an information network composed of the connections among routers and/or devices. We aim to use computing resources more efficiently and to improve network performance (delay time, communication bandwidth, and fault tolerance) by a proposed method of constructing a computing network with the characteristics of the brain network.
We focus not only on the structure of brain networks but also on ways that perception functions in the human brain. Humans can use ambiguous and uncertain information to make situationally appropriate decisions to a high degree of certainty. An example of applying this cognitive mechanism to technology would be a video player that can use cognitive functioning to automatically select an appropriate quality for display images according to the network environment, thereby improving the quality of experience (QoE) for the viewer. In addition, we are conducting research to estimate the degree of satisfaction (QoE) for watching videos by measuring electroencephalogram (EEG) of people who are watching the videos. Through the EEG measurement, we aim to realize network control technology that more closely matches the human experience by using technology to estimate QoE from human biological information. ( About Experiment (PDF)) We conduct research that applies findings from the field of brain science to research in the field of information networks.
Have you encountered MR? MR technology merges real and virtual worlds by overlaying virtual world information on the real world in real time. Though most current MR applications run offline, it is expected that they will connect to networks in the future. So-called 5G networks are being built in anticipation of increasing use of services, such as MR, that demand high bandwidth and real-time processing. The main features of a 5G network are multi-connection ability, high speed, high capacity, and low delay. Edge computing is one technology used in achieving these aims. Edge computing is a new form of information processing that is expected to work in tandem with cloud computing. For this pairing, a new processing base (an "edge server") is placed near users, such as at a mobile phone base station, and the edge servers, rather than a cloud data center, provide service to users. This structure can reduce delay caused by geographical factors and load concentration, which are problems in traditional cloud computing.
In the shopping mall experience service that we are developing as an implementation of network-oriented MR, a user wearing an MR headset cooperates with a robot in a shopping mall by video streaming or gesturing. Users can shop from home as if they were actually there. Since it is necessary to process the video taken by the robot in real time, we are considering the use of edge computing to improve the responsiveness of the application and raise the user's QoE.
Recently, devices intended for network use (IoT devices) have been increasing in number and are demanding more varied network services. IoT devices send private information to a user via the global network (e.g., a network camera sends recorded video, and a connected home appliance sends usage information specific to the user). New network services transmit important information (e.g., virtual currency services send monetary information and online shopping services send personal information). As the importance of sending information via networks increases, so does the importance of network security. Attacks on the network have already been observed and caused harm. As one example, an attacker could operate vulnerable IoT devices by sending packets via the network or could steal private information by infecting user terminals with malware distributed via malicious websites.
Our goal is to detect anomalous traffic by machine learning. We deployed a method that uses convolutional neural network (CNN) that learns the characteristics of malicious traffic by using known malicious traffic, which allows detecting malicious traffic from among actual traffic with only a low risk of false detection. We are also trying to construct a secure communication platform for IoT by using blockchain technology.
Network traffic can change markedly over time, and many methods have been proposed for responding to these changes. These methods involve changing the network structure, resource allocation, and the location of virtual network functions by using optical network technologies and network virtualization technologies, but rely on the assumption that a network controller can accurately obtain information about current traffic and predict future traffic. This introduces uncertainty because traffic predictions will include prediction errors and short-term traffic information for a large network is difficult to measure due to the overhead of monitoring and collecting the information.
To improve network control, we have developed methods that can function with inaccurate or partial information for each control interval. To develop the methods, we apply the process that allows the human brain to make many good decisions in highly uncertain environments.
Many people might find "virtualization" unfamiliar but would recognize the concept of a "virtual machine." Briefly, virtual machines run a software version of a computer on a physical machine. Virtualization thus uses software to simulate physical functions. The virtualization of networks is an active topic of study in the network industry. The data you transmit to the Internet is routed by specialized devices ("routers") until it arrives at the destination. The devices on the route can perform various network functions, including data processing. For example, firewalls seek to stop unauthorized access to data and anti-virus software seeks to detect viruses and other malware. In network virtualization, these routing and network functions are realized by software. Virtualization of the routing function is called software defined networking (SDN), and virtualization of the network function is called network function virtualization (NFV). Another important factor in virtualization is the controller. The controller centrally manages the software routers and network functions. Network virtualization and SDN/NFV technology development are globally advanced, and it is becoming possible to dynamically and flexibly control networks based on the actual conditions of the network.
We are particularly focused on devising algorithms for the controller. For example, we are working on a method of controlling the virtual network that is inspired by the "Yuragi" principle of living things, and on a control method that uses potential fields. We aim to create networks that can adapt to environmental changes such as traffic fluctuation and equipment failure.
The main functions and benefits are obtained by reconsidering the role of addressing. At present, Internet communications use so-called IP addresses. However, typical users are not conscious of the IP addresses used for communications. Most people have probably not considered them. For example, in the case of watching videos on YouTube, viewers care about the videos (i.e., the content), not the addresses of the YouTube servers (i.e., the IP addresses).
ICN has been proposed as a new-generation network architecture that addresses mismatches between the physical properties of the Internet and actual usage, with the aim of increasing the efficiency of handling content. Toward this, ICN uses the names of content as the routing information, rather than using IP addresses. For example, with ICN, a target video could be retrieved by sending a request for the content by name only, such as "youtube/sport.mp4". Processes used now, such as resolving "www.youtube.com" with a domain name server to retrieve the IP address and then routing the request to YouTube and sending the desired video ID would not be necessary with ICN.
Moreover, using names with ICN offers many other benefits. As one example, when IP addresses are used for routing, the routers are unaware of communication details. ICN, in contrast, allows knowing the details of content from the content name. If the same content, decided by name, is requested again, the content can be treated as a duplicate. Therefore, if routers keep a copy of content and reply with a copy upon receipt of a request for that content, communications can be more efficient by shortening the routing. This is just one example of ICN's benefits. There are many additional benefits, including supporting multicasting as standard, improving mobility, and strengthening security. In our laboratory, we set the goal of problem-solving for the realization of ICN, and we research various possibilities, such as hardware architecture, routing architecture, drone-based movable routers (shown in the figure) and other new ICN services. We are engaged in this research to realize ICN and design future communications.
Further information on the research of our laboratory can be found in our annual research reports. Each academic year, we organize our research achievements and provide a list of our published papers. In the report, the research background, the research topics, the list of published papers, and the research staff for the academic year are listed in Japanese. Many papers in the list are written in English.