Professor Yukihiro Bando participated as a co-author when he was at NTT, and his research work with Takamoto Sasaki and others was recorded in the paper magazine IEEE Access (Impact Factor 3.6).

https://ieeexplore.ieee.org/document/11173648

In this paper, we propose a high-speed, parallelizable specific value threshold processing method for efficient low-ranking analysis of a large number of small matrices as a technology that contributes to the generation of high-quality 3D point groups essential for digitalization in the real world. In conventional low-rank analysis, the iteration of operations called "specific value threshold processing" was key to high quality, while the high computational load became a bottleneck in processing, limiting practicality.

In this study, we newly discovered the geometric characteristics of "nuclear norm", which is the essence of low-ranking structures, and derived a computational process that is fundamentally different from conventional methods. By establishing a new processing method that does not require specific value degradation (SVD), which was a bottleneck of conventional methods, we have achieved faster calculations and parallelization at the same time. As a result, the proposed method achieves high-speed processing of up to 60 times more than conventional methods, while maintaining high computational accuracy.

These results can be applied to processing such as noise removal for three-dimensional point clouds, and are expected to be an important factor in the future realization of cyber-physical systems (CPS).

[Contact information]
Shimonoseki City University
Corporate Planning Department Public Relations and Branding Strategy Division
TEL. 083-253-8967