At the boundary between data science and computer networks, we are working to create innovative technologies for computer networks using state-of-the-art machine learning, optimization, and cryptography. Aiming to realize Society 5.0, a human-centered society that balances economic development with the resolution of social issues, it is becoming increasingly important to collect, analyze, and utilize data from a vast array of sensors through systems that highly integrate cyberspace and physical space. We conduct research on innovative technologies for architecture and operation of edge computing and cloud computing systems. In particular, we have recently focused on data analysis and data distribution. In data analysis, we are working on traffic prediction methods and traffic anomaly detection methods, and in data distribution, we are working on edge-cloud data distribution architectures and distributed information exchange infrastructures.
We are working on a method for predicting the amount of communication traffic from GPS devices. Using Bayesian networks, we will develop a method for predicting complex variations in communication traffic in space and time, which has been difficult to predict using conventional methods.
To cope with ever-changing attack patterns, we implement a communication traffic anomaly detection method using semi-supervised machine learning in order to reduce the work required to label training data for re-training.
Develop an IoT platform connecting edge computers and data analysis servers to establish a foundation for collecting large amounts of sensor information. Optimization of chip architecture for machine learning inference at edge nodes will be studied. In addition, to ensure that the exchange of machine learning data between edge and cloud nodes is completed within the deadline, we will implement deadline-aware data transfer scheduling using reinforcement learning.
Until now, centralized certificates have been required for information distribution on digital infrastructures, and management costs and confidentiality have been major problems. To solve this problem, we use blockchain technology to realize a data distribution infrastructure in which certificates can be authenticated in a decentralized manner. Furthermore, we will realize machine learning and federative learning using data that is encrypted using secret computation of fully homomorphic encryption.