HTML5 Webook

29/64

NICT REPORT 27voice resources, bio-related data, brain-relat-ed data) and publishing them on the AIS website. We have also conducted a wide range of surveys for user evaluation in NICT and to extract very scarce data.Neural big data infrastructureIn the neural big data infrastructure project, we are collaborating with the Cen-ter for Information and Neural Networks to prepare an R&D system where the large-scale collection of fMRI/MEG brain activity data and AI technology such as machine learning are used to implement next-gen-eration AI systems for the analysis and simulation of brain activity, including brain biomarkers (Fig.4), brain information de-coding, an electroencephalogram testbed, and brain & sports/wellness.Accelerating the study of neural translation dedicated to patent documentationBy entering into an agreement with the National Institute of Advanced Industrial Sci-ence and Technology on promoting collabo-ration and cooperation in the information communication field, we have started a joint study based on this agreement on the subject of neural machine translation dedicated to patent translation and a system structure that makes this possible. We conducted research on the use of multiple GPUs to perform paral-lel acceleration of neural machine translation training, and achieved speeds approximately four times faster than were possible with a single GPU.Operating a translation bankAt NICT, we are researching and develop-ing neural machine translation as part of a Global Communication Plan aimed at provid-ing society with multilingual speech transla-tion technology by 2020. Improvements to the algorithms of neural networks can be an effective way of improving the accuracy of neural machine translation, but it is vitally im-portant to securing large quantities of transla-tion data from various fields.For this reason, in September 2017, we started operating a nationwide translation bank system (Hon’yaku Bank) to collect translation data in collaboration with the Min-istry of Internal Affairs and Communications. We launched a website to raise awareness and improve people’s understanding of this service, and in order to offer some merit for providing this system with translation data, we prepared a system whereby, when calcu-lating the user license fees for NICT’s ma-chine translation technology, the charges are reduced in consideration of the estimated amount of translation data provided by a user. In the future, it is expected that Japan’s trans-lation technology will become more accurate and cover a wider range of fields due to the use of translation data collected by NICT via this translation bank (Fig.3).During the period of roughly six months following the launch of the translation bank, over 50 companies have signed up to par-ticipate in the project.AI × securityIn recent years, cyberattacks have posed new threats due to the accelerated use of AI and automation based on machine learning and the like, and there is an urgent need for more automation and AI in the security op-erations that deal with these attacks. For this reason, we launched the AI × security project in cooperation with the Cybersecurity Re-search Institute, and in FY2017 we identified specific tasks and built a research and devel-opment system (Fig.5).Participating in discussions on AI-related policy proposals, etc.We are cooperating with external organi-zations by taking part in discussions of AI-re-lated policy proposals at COCN (Council on Competitiveness–Nippon) and DiTT (Digital Textbook and Teaching), among others.Fig.3 : Our translation bank concept (Hon’yaku Bank)Fig.5 : Overview of AIS cybersecurity projectFig.4 : Development of software for measuring cognitive function with a shorter test