Project
Graph Data Analysis and Management
Graph data can model relationships between entities and is used in various familiar applications. For example, structuring relationships between entities as knowledge graphs enables their use in web search and product recommendation systems. Similarly, molecular data can represent atoms as entities and the connections between them as relationships, allowing applications such as searching for molecules with similar data structures. Social networking services and road networks are also closely related to our daily lives, and graph data is becoming increasingly large and diverse.
We are conducting research and development on graph data from four perspectives: management, search, discovery, and prediction. Specifically, we are researching the following technologies:
- Database technologies that enable efficient management and high-speed search
- Data mining techniques for discovering new insights
- Deep learning techniques for accurate predictions
For example, we are developing technologies for speeding up queries in graph databases and extracting the strength and characteristics of relationships in graph data to evaluate whether exceptional relationships or discriminatory biases are present. Additionally, we are developing deep learning techniques applicable to large-scale graph data and methods for automatically selecting appropriate graph deep learning models.
- I am closely collaborating with the Technical University of Eindhoven (Netherlands) and the University of Copenhagen (Denmark).
- I am the only academic institution from Japan participating in the Linked Data Benchmark Council (LDBC), which aims to standardize graph databases (as of March 2023, there are two members from Japan).
- I am conducting research on the theme "Creation of fundamental technologies for explainable bias in graph data" in the JST PRESTO "Trusted AI" area.
- I have published a website for Graph Neural Network (GNN) learning called "Graph Dojo" and a GNN tutorial video.
- I oversee the international collaborative research project Musubi.
Trustworthy Artificial Intelligence
Artificial intelligence has become widespread in society and is being used for decision-making. In recent years, it is not only about achieving high accuracy but also about how much trust it can gain from humans, which has become a critical topic.
To realize trustworthy AI, we are conducting research on fairness to prevent AI from reinforcing discrimination, explainability to interpret prediction results, and federated learning techniques to protect personal data. We also study the trustworthiness of AI from perspectives beyond computer science, including philosophy and psychology.
- We are conducting research under JST ASPIRE on the theme "A comprehensive framework for fair graph data analysis."
- We have published a federated learning tutorial video.
Mobile and Spatiotemporal Data Analysis and Management
With the advancement of the Internet of Things and location-based services, much data now includes time and location information. Examples include human flow, traffic data, restaurant information, and sensor data. Today, many people use services involving spatiotemporal data without even realizing it.
We are also conducting research and development on mobile and spatiotemporal data from the perspectives of management, search, discovery, and prediction. For example, we are developing technologies for speeding up spatial queries (such as finding data within a range or the nearest data point), optimizing distributed parallel processing, mining relationships between data, and predicting sensor data values using deep learning.
Additionally, in mobile and spatiotemporal data analysis, graph data analysis techniques can be applied to capture correlations based on road networks and distances.
- We are working closely with Shanghai Jiao Tong University and applying spatiotemporal data analysis techniques to urban engineering.
Cross-Disciplinary Applications of Data Science Technology
Data science and artificial intelligence technologies are spreading widely throughout society. Even fields that have not traditionally utilized data science technologies are now engaging in related research, and we are conducting joint research with many researchers from different disciplines.
For example, we are collaborating on discovering new useful substances in chemistry and materials science, analyzing and predicting data for patients with urinary tract stones in medicine, examining social issues of AI in philosophy, and conducting traffic analysis and energy demand forecasting in urban engineering. By applying core information processing technologies to a wide range of fields, we can acquire new knowledge beyond informatics and grasp the technologies currently needed by society, contributing to the development of research ideas and new technologies.