HTML5 Webook

21/64

NICT REPORT 19analysis. NIRVANA-KAI (NICTER Real-net-work Visual Analyzer KAI) is a security platform against advanced persistent threats (APTs). It collects and analyzes se-curity alerts provided by various security appliances and end hosts, and it automati-cally deploys new controls to the appli-ances to prevent APTs (Fig.1). This security platform, in combination with other secu-rity appliances, visualizes security-related events (e.g., cyberattacks) in real time and thus facilitates the implementation of prompt and adequate measures. We have been successfully transferring the technol-ogy of NIRVANA-KAI to Japanese industry since 2015. Security testbed development and operations technologyWe oversee R&D on technologies for emulating cyberattacks in a safe environ-ment. In particular, we are developing a security verification platform for verifying new protection technologies and counter-measures in an emulation environment. In this security testbed development, we are investigating a large-scale deception framework called STARDUST for luring hu-man adversaries with the aim of attributing sophisticated cyberattacks such as APTs (Fig.2). STARDUST can quickly and flexibly build mimetic enterprise networks called ‘parallel networks.’ In a parallel network, APT malware can be executed and ob-served in a highly stealthy manner. A wormhole connects parallel and real net-works so that the parallel-world network can pretend to have the same IP address-es as the real-world network. STARDUST enables us to stealthily observe adversar-ies’ activities on parallel networks and to feed-forward the findings of the observa-tion to APT countermeasures such as NIRVANA-KAI to promote their evolution. Cryptographic technologiesOur R&D on functional cryptographic technologies provides new functionalities to meet the evolving social needs accom-panying the growth of IoT and to evaluate cryptographic technologies. We are con-tributing to the standardization of new cryptographic technologies and to the construction of safe and secure ICT sys-tems. We also engage in R&D on privacy-enhancing technologies aimed at the safe utilization of personal data and technical support for appropriate privacy measures. Recently, deep learning has gained con-siderable attention in both industry and academia. Although the collection of mas-sive amounts of data is vital for deep learn-ing, it raises the issue of privacy. To resolve this issue, we are building a privacy-pre-serving deep learning system, where many learning participants perform neural-net-work-based deep learning over a com-bined dataset of all participants without revealing the participants’ local data (Fig.3). This privacy-preserving deep learn-ing system allows us to integrate isolated and hidden small data as big data. One of our promising technologies for realizing the system is ‘homomorphic encryption,’ which allows computations to be carried out on encrypted data without decryption. We are investigating the application of this system to a broad range of fields, such as medical genetics and illegal remittance detection, in a privacy-preserving manner. Fig.2 : STARDUSTA ‘parallel network’ imitating an organization network can be built on the STARDUST platform to at-tract attackers. It can observe the behavior of an attack on the network.Fig.3 : Privacy-preserving deep learningMany learning participants perform neural-network-based deep learning on a combined dataset of all participants without revealing the local data of individual participants.Research and Development