Japan Science and Technology Agency (JST)
JST-Mirai Program "Super Smart Society (Society 5.0)" mission area:
Prioritized Theme: Modeling and AI that Connects the Cyber and Physical Worlds
Engineerable AI Techniques for Practical Application of High-Quality Machine Learning-based Systems
Principal Investigator: ISHIKAWA, Fuyuki (Associate Professor, Information Systems Architecture Science Research Division)
High expectations are placed on the application of Artificial Intelligence (AI) in various fields because it can find regularities hidden within a large amount of data, which can then be used for classification, predication, and anomaly detection. However, since conventional AI requires a large amount of data for training, it is difficult to deal with cases in which only a small amount of data can be obtained or where modifications are required. For example, when AI is used in the healthcare field, a problem is likely to arise since it easily overlooks atypical lesions, for which it is difficult to obtain a large amount of data. In addition, when errors have occurred in road sign recognition in autonomous driving using AI, it takes an enormous amount of time to correct the errors. In order to apply AI to critical fields like healthcare and autonomous driving, the solving of these issues becomes an urgent task. We are therefore aiming to establish a new general-purpose fundamental technology called "Engineerable AI (eAI)" to build up and enhance the safety and reliability of AI. In contrast to conventional AI based on learning and repetitive correction through the use of large amounts of data, eAI is a technology that guarantees and corrects AI operations by extracting and analyzing not only the technologies used in constructing AI that reflect human knowledge, but also factors that cause AI errors.
This research and development project is expected to bring into realization a diagnostic support system that can detect atypical lesions, even with a limited amount of data, thus contributing to alleviating the shortage of medical specialists and rectifying the irregular quality levels in healthcare. In the case of autonomous driving, the ability to extract and target specific AI performances that need to be corrected will reduce the time required for system development and contribute to the improvement and assurance of safety in autonomous driving. Our aim is to demonstrate the effectiveness of eAI in solving problems in healthcare and autonomous driving and contribute to the establishment of internationally competitive production technology incorporating eAI.