Mathematical Modeling and Deep Learning for Small-Data AI

• Overview

  Current technologies of artificial intelligence (AI), including deep learning, are effective as long as we have many high-quality data where their quality is often measured by the “4V” of Big Data-Volume, Variety, Velocity, and Veracity. In contrast, we often encounter “Small Data” in physical space such as pandemic of disease spreading over social networks, biomedical information like EEG and fMRI, and sensor signals collected by numerous Internet-of-Things (IoT) devices. Discovering knowledge and extracting useful information from such Small Data are highly required to integrate cyber and physical spaces that results in a reliable and stable cyber-physical system. In this project, we study “Small-Data AI” that unifies mathematical modeling and deep learning from signal processing and machine learning perspectives.

  

• Approaches

  In this project, we study both theoretical and practical aspects of Small-Data AI. Its effectiveness will be validated through real-world experiments in biomedical information processing and IoT data analysis. Our international team has its strengths both on theory and applications in the project and tackles the investigation of reliable and robust AI technologies.

  The analysis group in the project studies theory and algorithms of signal and information processing for epidemics, biomedical data, and IoT signals. The application group works on various aspects on Small-Data AI: Diagnostic imaging, brain computer interface, and IoT-based robotics.

  Researches on Small-Data AI have been conducted in various fields. However, there are few studies that aim to theoretically unify these researches. This project will shed light on signal processing as a fundamental theory of Small-Data AI.

  

• Plan

  This project composes of two groups, i.e., the analysis group and the application group. Both groups interact with each other. The analysis group works on theory and algorithms of Small-Data AI and the application group studies its applications on biomedical information processing, robotics, and IoT/sensor networks. Real-world problems are studied through the collaboration of both groups that results in generalized and flexible theory/algorithms.

  TUAT has its strength on AI technologies based on signal processing and machine learning. We invite world-renowned researchers in this field for accelerating the progress of this project.. Collaborating with them will revolutionize future Small-Data AI including medical and robotics technologies.

  

Members

Toshiyuki Kondo 　(Institute of Engineering / Professor)
Akinobu Shimizu 　(Institute of Engineering / Professor)
Hiroshi Ishida 　(Institute of Engineering / Professor)