Spatio-Temoral Pattern Mining

Natural phenomena, biological activities, and modern devices consist of multiple elements that maintain order through their interactions. With the goal of understanding the mechanisms of multi-dimentional systems or prediction, research in spatio-temporal pattern mining is conducted to extract causal patterns of events from discretely occurring event sequences. As a specific application, analyses have been performed on damage patterns from sequences of ultrasonic signals resulting from fuel cell damage, as well as the analysis of interactions between earthquakes using earthquake source list data.

Sleep Assessment by Machine Learning

With the aim of daily sleep quality evaluation, we are researching machine learning methods for a convenient sleep assessment based on "sounds" during sleep. The sounds contain various physiological activities during sleep, such as snoring, teeth grinding, body movements, etc. By analyzing the complex sounds using deep learning, we are researching techniques to discern sleep quality, as well as identifying its factors. Also, we have constructed a large sleep database in home environments with data from hundreds of individuals, and our research also involves multimodal learning that considers not only sound but also environmental, physical, and psychological factors.

Equation Discovery from Observation Data

Various physical phenomena, including climate, are modeled by partial differential equations that describe their behavior in both time and space. If we can estimate the governing equations of the phenomenon from observational data, it can be valuable for understanding and predicting unknown phenomena. Therefore, we are researching machine learning methods to exploratively discover governing equations from observational data. A pioneering approach in this field is the Physics-Informed Neural Network (PINN), which incorporates equation-based constraints into neural networks. In this study, we are extending the PINN to the discovery of equations for a broader range of phenomena.