Student Research Programs

Sponsors in 2026 include the following:

PROJECT 1: Mitsubishi: Integrated Computational Modeling for Accelerated Single-Pixel Imaging

Single-pixel imaging (SPI) is a technique that reconstructs images from measurements using modulated illumination and a single-pixel detector, enabling high-sensitivity and low-cost imaging even beyond the visible spectrum or under extremely low-light conditions. However, it generally requires sequential projection of a large number of modulation patterns, making acquisition speed a bottleneck. Introducing unsupervised deep learning into SPI reconstruction can reduce the number of required samples and accelerate recovery, but the ill-conditioned nature of the inverse problem derived from the measurement equation, coupled with the absence of direct supervisory signals in unsupervised learning, may degrade reconstruction accuracy. In this project, we investigate an optimization approach that integrates mathematical regularization and physical models into deep learning to reduce performance. The proposed method will be evaluated using reconstruction accuracy, robustness to noise variations, and acquisition speed as key metrics.

More details about the project can be found here.

PROJECT 2: Fujitsu: Mathematical and data-driven approaches to materials sciences

In the fields of materials sciences, advancements in both experimental techniques and simulation techniques have significantly enhanced the quality of data, resulting in an era of unprecedented data richness. NanoTerasu Synchrotron Light Source, which began operation at Tohoku University in 2024, enables the measurement of molecular, atomic, and electronic states with nanometer-level high spatial resolution. AI-assisted molecular dynamics simulations have successfully reproduced high-precision motions and chemical reactions of large-scale systems exceeding 100,000 atoms.

One goal of modern materials sciences is to develop new functional materials to drive innovation and contribute to resolving societal issues, including environmental challenges. As experiment and simulation performances improve drastically, the amount of data created increases accordingly. One big challenge here is to exhaustively extract useful and exciting information from such a huge amount of experimental and simulation data without the bias of human experiences or intuitions in order to finally lead to new discoveries. While computing power and AI have been playing an important role to solve the issue, the power of mathematics to clarify the essence of things is now essential.

In this project, you will develop methods for analysis of data in the field of materials sciences assisted by mathematics.

More details about the project can be found here

PROJECT 3: Daikin: tbd

Please check back in a few days for more details about this project.