Because we have scientific knowledge cultivated to this day, should AI just learn data? We are engaged in research that integrates existing scientific methods, which explain the composition of data with mathematical models expressing human knowledge, and the new methods of data science, as represented by deep learning. For example, a doctor must be able to explain the results of a data analysis that support the diagnosis. In such a case, data science grounded in human knowledge will be beneficial.

Underpinning the dramatic development of information technology in modern society is the sharp rise in the performance of computers themselves. However, going forward the conventional approach of simply extending computers’ performance as in the past will be difficult. We are conducting research on realizing “soft” computers in which logical flexibility is introduced into their hardware structure. At the same time, we are also actually creating these computers.

Audio user interfaces like speech recognition are technologies that anyone can easily use. We are conducting integrated research on speech and language with the goal of creating man-machine interfaces using speech recognition and comprehension and dialogue systems. We are also conducting research on bioacoustic signal processing to identify, for example, healthy patients and patients with pulmonary illnesses using respiratory sounds obtained by auscultation and to infer information about an infant’s condition using his or her crying sounds.

As information devices become more advanced and improve in performance, technologies such as VR (Virtual Reality), AR (Augmented Reality), and MR (Mixed Reality) have become commonplace. These technologies can be used easily by anyone. We are conducting research on applied technologies using VR/AR/MR to support domains like rehabilitation medicine and tourism. In addition, due to the expansion of the Internet, the illegal distribution of images and video over networks has become an issue. We are researching digital watermarking as a solution to this problem.

We are developing mathematical models and algorithms for searching for genes relevant to predicting diseases from genomic data. First, strongly relevant genes are narrowed down from genomic data (1). Using the discovered genes (2), predictive modeling is performed (3). The development of algorithms makes use of extensive knowledge of mathematic statistics, biostatistics, genetics, and machine learning.

The figure below shows an information platform developed in cooperation with Nagasaki Prefecture and Nagasaki City for collecting and analyzing tourism-related big data. This system provides information for quick visualization, such as where tourists are staying and how long they stay in spots, in order to understand tourism trends. The information is based on smartphone log information and information available on the Web.

The figure below shows artificially created scatter plots of relationships between curved class grade (y), study time (x1), and height (x2). y seems proportional to x1 and independent of x2. Making these determinations by automatically judging from data is called model selection. The research theme here focuses on statistical inference, which includes exploring the significance of some extracted variables after carrying out model selection.

We are tackling the mystery of Egyptian pyramids, built about 4,500 years ago. We know almost nothing about these giant buildings constructed by humankind. How were the stones assembled? What were inside the pyramids? And, how were they designed? For ancient Egyptians, the design and construction of pyramids solved an unknown problem. Unraveling this mystery by humanity today will offer important clues for solving our own yet-unknown problems.