Fumiko Ogushi

Characterizing the collaborative degital knolwdge space

Researcher

Fumiko Ogushi

Outline

Collaborative digital knowledge space like Wikipedia is organized in a collaborative and bottom-up way with voluntary contributions and improved its quality of the information in a self-organized manner. Thus, the lack of selective professional writers and editors makes the judgement about the quality and trustworthiness of the information a real challenge. This study aims to elucidate the characteristics of self-organized and complex structure of collaborative digital knowledge space and establish the metrics for quantifying the quality and importance of its constituents.

Yasushi Kawase

Modeling and algorithms in multiagent environments

Researcher

Yasushi Kawase

Outline

In this project, we will develop efficient algorithms that compute high-quality solutions while taking strategic issues into account for essential problems in multiagent environments, such as stable matching problem, fair division problem, and multidimensional auction problem. Furthermore, we construct fundamental technologies for modeling and algorithm design methods for multiagent environments based on the obtained knowledge.

Akiyoshi Sannai

Research on deep learning using symmetry and related invariant theory

Researcher

Akiyoshi Sannai

Outline

We are currently trying to adapt the invariant deep neural network model that we were able to construct using Reynolds operators to the invariant deep neural network framework. We are developing the theory of Reynolds dimension and Reynolds design, which are necessary mathematical frameworks to construct the invariant deep neural network model by approaching from both mathematics and information.

Mitsuru Shibayama

Orbital design of artificial satellites and planetary probes by variational and geometrical methods

Researcher

Mitsuru Shibayama

Outline

In recent years, space development has been in the limelight. Until now, rocket orbit design has mainly used the solutions of the two-body problem or the orbit obtained by numerical calculation of the restricted three-body problem. In this study, I will construct a novel trajectory by an approach that has not been done so far by applying a mathematically advanced theory.

Sho Sonoda

Theory and application of depth structure inherent in geometric datasets

Researcher

Sho Sonoda

Outline

Deep learning has revealed the effectiveness of the method of decomposing a map in the depth direction (i.e. composite map), or the “depth decomposition”. The method of decomposing a map in the width direction (i.e. basis and coefficients) is ubiquitous in information technology and has been well-organized as harmonic analysis. On the other hand, the theory of depth decomposition is still incomplete, and the properties of the intermediate information representation obtained by deep learning are largely unpredictable. In this research project, I develop the theory and methods of depth decomposition, aiming to formulate the “depth” of mappings and data in particular, and deploy them to the next generation of information technology.

Shinichi Tanigawa

Combinatorial Analysis in Computational Geometry

Researcher

Shinichi Tanigawa

Outline

Understanding configuration spaces in geometric constrained systems is one of the major topics in computational geometry. In this project, we explore combinatorial properties behind generic geometric constrained systems based on techniques in combinatorics and combinatorial optimization. The research aims to develop a novel combinatorial foundation of interpretable algorithm design.

Yohei Hosoe

Model Based Adaptive Learning Control Using Probabilistic and Statistical Information

Researcher

Yohei Hosoe

Outline

In model based control, the quality of a mathematical model directly affects control performance. Depending on the plant, however, it may not be possible to obtain a sufficiently accurate model. This project develops theory and techniques that enable us to complement the unknown part of the model with a distribution of stochastic processes by using priori and posteriori information, and utilize it for control. Through this development, this project aims at laying a foundation for solving social issues related to the automatic control of various plants.

Kei Majima

Quantum inspired machine learning for high dimensional neural data analysis

Researcher

Kei Majima

Outline

Machine learning algorithms specialized for neural decoding have allowed the extraction of information encoded in the brain. However, their application to high dimensional data is limited due to their large computational complexity. To solve this problem, we develop scalable machine learning algorithms by using computational techniques developed in the field of quantum computation.

Yuto Miyatake

Uncertainty quantification for numerical integration of evolution equations

Researcher

Yuto Miyatake

Outline

Numerical calculation of evolution equations plays an essential role in modern science. In recent years, the increase in demand has been more remarkable than the progress of algorithms and computers, and thus there is a strong need for a method evaluating the reliability of numerical computations quantitatively. However, little effort has been paid for that purpose. This project aims to develop practical uncertainty quantification methods and establish fundamental theories by combining probabilistic and statistical viewpoints with the knowledge of numerical analysis.

Yoh-ichi Mototake

Mathematical Structure Extraction of Pattern Dynamics by Interpretable AI and Its Application to Materials Informatics

Researcher

Yoh-ichi Mototake

Outline

It is important to develop a better understanding of pattern dynamics, which is found in various fields of materials science, to predict the time evolution of material formation and fracture processes in complex environments. In this research project, I will develop a framework for extracting the interpretable mathematical structures of pattern dynamics by integrating machine learning methods such as topological data analysis, deep neural networks, and Bayesian inference. It is expected that the results of this research project will facilitate scientists’ efforts to explore extrapolable general principles of pattern dynamics.

Yu Yokoi

Investigating and utilizing the structure of matchings under preferences

Researcher

Yu Yokoi

Outline

The theory of matching under preferences has developed significantly in recent years. The purpose of this area is to design methods to compute efficient and fair matchings based on people’s preferences. In this project, I investigate generalized matching models that can express various preferences and analyze the structure of the set of desirable matchings. Exploiting the obtained structural results, I aim to design matching algorithms that achieves fairness and optimality, while taking into consideration the strategic behavior of participants.