Energy-efficient information processing
For realizing next-generation information processing systems, it is indispensable to miniaturize devices and make structures compact for enhancement of energy efficiency. We aim to develop mathematical methodologies for making brain-like computing systems energy efficient such that efficient computing is realized with low power and high speed. Examples of our studies are as follows:
- Reservoir computing ([Review])
- Energy-efficient associative memory neural networks ([Paper])
Applications of machine learning and advanced mathematical methods
Machine learning technologies have enabled to efficiently perform tasks that have been manually handled by people. We aim to mathematically formulate problems in fields that are not approached by machine learning and mathematical modeling, and solve the problems by combining appropriate machine learning methods and advanced mathematical techniques. Examples of our studies are as follows:
Network robustness
Networked systems are ubiquitous in the world, such as the Internet, power networks, and biological networks. Networking often accompanies a risk that a partial failure causes a breakdown of the whole system. We are investigating how network robustness depends on network structure, dynamics, and element interactions. Our aim is to develop a design method of robust networks and a recovery method of damaged networks. Examples of our studies are as follows:
- Robustness of complex oscillator networks ([Paper])
- Robustness of biological networks ([Book chapter])
- Targeted intervention to epidemic spreading ([Paper])
Fundamental theory of complex systems dynamics
We are developing fundamental methodologies for understanding and analyzing complex systems. Depending on the condition, system behavior can change from order to disorder, from simple to complex, and from regular to irregular. We aim to understand the mechanism of such qualitative changes of system behavior.