[Trusted quality AI systems] Year Started : 2022

Atsushi Shimada

A Development of Reliable Learning Analytics Platform and its Advanced Applications for Teaching and Learning Support

Grant No.：JPMJCR22D1

Research Director

Atsushi Shimada

Professor
Faculty of Information Science and Electrical Engineering
Kyushu University

Collaborator

Daisuke Deguchi	Associate Professor Graduate School of Informatics Nagoya University
Takayoshi Yamashita	Professor College of engineering Chubu University

Outline

Learning analytics (LA) is research that provides effective support for teaching and learning based on analysis of educational data. This project will develop a new learning analytics platform, named “ReLAX”: Reliable Learning Analytics for X, that realizes immediacy to quickly provide analytics results, persuasiveness by making the analysis processes transparent, and generalizable analytics methodologies for various educational data.

Shohei Shimizu

Causal discovery and its applications for reliable AI

Grant No.：JPMJCR22D2

Research Director

Shohei Shimizu

Professor
Faculty of Data Science
Shiga University

Collaborator

Kei Inoue	Researcher Clinical Development Department SUSMED, Inc
Takehiko Hayashi	Chief Senior Researcher Social Systems Division National Institute for Environmental Studies
Shingo Fukuma	Associate Professor Graduate School of Medicine Kyoto University

Outline

Statistical causal inference using causal graphs is essential in improving the explainability, fairness, and performance needed for reliable AI. To perform statistical causal inference, a causal graph needs to be drawn by the analyst, but it often is the case that insufficient domain knowledge is available for this purpose. Then, a causal structure search methodology that uses data to infer a causal graph, i.e., causal discovery, is useful. Thus, we develop methods for causal discovery to infer causal graphs from data and use them to analyze explainability and fairness in the four areas of policy science, environmental studies, preventive medicine, and clinical medicine.

Mahito Sugiyama

Machine Learning That Connects to Symbolic Reasoning

Grant No.：JPMJCR22D3

Research Director

Mahito Sugiyama

Associate Professor National Institute of Informatics Principles of Informatics Research Division

Collaborator

Katsumi Inoue	Professor National Institute of Informatics Research Organization of Information and Systems
Ryosuke Kojima	Lecturer Graduate School of Medicine Kyoto University
Masaaki Nishino	Distinguished Researcher NTT Communication Science Laboratories NTT Corporation

Outline

To develop fundamental methodologies of trustworthy AI, we aim at unifying modern machine learning approaches with potentially massive parameters and symbolic reasoning that enjoys high interpretability through the geometric lens. We design and construct machine learning systems inherently connected to symbolic reasoning, which simultaneously solve the problems of the reliability of machine learning and the robustness of symbolic reasoning.