Engineerable AI Techniques for Practical Applications of High-Quality Machine Learning-based Systems
|Prioritized Theme||Modeling and AI that Connects the Cyber and Physical Worlds|
Machine learning techniques enable to implement functionality that satisfies a fuzzy goal in an inductive way by training from data. However, this approach makes it impossible to handle rare situations or to understand and fix performance limitations over many situations. This characteristic poses an obstacle in application to safety- and reliabilitycritical domains.
We envision techniques for "Engineerable AI" that support development, quality assurance, and operation of AI systems, thus promoting practical applications of high-quality AI systems. We develop techniques for construction of AI by incorporating human knowledge into deep learning and for assurance and improvement by extracting and analyzing factors affecting the target quality. We promote this research by collaboration of the machine learning area and the software engineering area and demonstrate the effectiveness in autonomous driving and healthcare.
* This project is launched by integrating the following feasibility studies;
"Design of Data-driven Hierarchical Supervisor Using Formal Methods" (USHIO Toshimitsu), "Development of a new deep-learning model that can handle both images and symbolic data and studies on its interaction with humans" (SUZUKI Kenji) and "Value Creation by Reliable Machine Learning-based Systems" (YOSHIOKA Nobukazu)
Affiliation and job title should automatically appear from the information that a researcher registered with researchmap.
Data may be outdated or undocumented.|
When there is not a connection via the internet, data are not displayed.