研究業績リスト
その他
作成日時 01/04/2025–31/03/2028
Offer Organization: 日本学術振興会, System Name: 科学研究費助成事業, Category: 基盤研究(B), Fund Type: -, Overall Grant Amount: - (direct: 14400000, indirect: 4320000)
その他
Establishment of Computational Methods for Ultra-Large-Scale Matrix Functions
作成日時 01/04/2025–31/03/2030
Offer Organization: Japan Society for the Promotion of Science, System Name: Grants-in-Aid for Scientific Research, Category: Grant-in-Aid for Scientific Research (S), Fund Type: -, Overall Grant Amount: - (direct: 103200000, indirect: 30960000)
その他
作成日時 04/2023–03/2024
Offer Organization: -, System Name: 2023 年度TIA 連携プログラム探索推進事業「かけはし」, Category: -, Fund Type: -, Overall Grant Amount: - (direct: -, indirect: -)
その他
組合せ的前処理と量子アニーリングの融合による行列計算の加速手法
作成日時 30/06/2022–31/03/2025
Offer Organization: 日本学術振興会, System Name: 科学研究費助成事業 挑戦的研究(萌芽), Category: 挑戦的研究(萌芽), Fund Type: -, Overall Grant Amount: - (direct: 4600000, indirect: 1380000)
その他
高いスケーリング性能と高精度性を併せ持つ次世代固有値・特異値分解ライブラリの開発
作成日時 01/04/2019–31/03/2023
Offer Organization: -, System Name: -, Category: -, Fund Type: competitive_research_funding, Overall Grant Amount: - (direct: 7840000, indirect: 2352000)
その他
Fast, accurate and stable matrix computation algorithms based on non-orthogonal transformations
作成日時 30/06/2017–31/03/2022
Offer Organization: Japan Society for the Promotion of Science, System Name: Grants-in-Aid for Scientific Research, Category: Grant-in-Aid for Challenging Research (Exploratory), Fund Type: -, Overall Grant Amount: - (direct: 4700000, indirect: 1410000)
The aim of this study is to develop new algorithms for matrix computations based on non-orthogonal transformations and analyze their convergence properties and accuracy theoretically. We chose the time-dependent eigenvalue problem and preconditioning methods for iterative solution of linear systems as examples. For the former, we developed an algorithm based on matrix multiplications, which is suited for next-generation microprocessors. For the latter, we developed a highly parallel algorithm based on the modified incomplete Cholesky factorization and the block red-block ordering.
その他
作成日時 01/04/2017–31/03/2020
Offer Organization: Japan Society for the Promotion of Science, System Name: Grants-in-Aid for Scientific Research, Category: Grant-in-Aid for Scientific Research (B), Fund Type: -, Overall Grant Amount: - (direct: 10100000, indirect: 3030000)
In this study, we developed matrix solvers which are applicable to a wide range of computer physics applications, such as electronic structure calculation, plasma simulation and crack growth simulation. Our main targets are direct solvers for sparse and band matrices and generalized eigenvalue solvers for dense matrices. With the use of communication-avoiding algorithms and automatic code selection techniques, our solvers aim at achieving high strong-scaling performance. We applied our solvers to time-dependent simulation of organic polymeric materials and crack growth simulation and verified their performance.
その他
作成日時 19/07/2016–31/03/2019
Offer Organization: Japan Society for the Promotion of Science, System Name: Grants-in-Aid for Scientific Research, Category: Grant-in-Aid for Scientific Research (B), Fund Type: -, Overall Grant Amount: - (direct: 13700000, indirect: 4110000)
In this research, we aimed at developing a new designing framework of innovative electronics material for which conventional designing approach has become too expensive due to the rapid increase of the complexity and size of target designs. The main idea there was to establish a unified approach where the physics, the mathematical model, and its numerical computation were demanded to be consistent in terms of a certain mathematical structure. As outcomes of this research, we have developed several important supporting new techniques, such that new structure-preserving model reduction techniques, new fast iterative solvers for electronics dynamics computations. Then we considered some electronics dynamics computations by efficient, energy-conserving parallel numerical methods, which confirmed the validity of the concept of the unified approach in that it actually enables efficient, stable computation.
その他
Theory and Application of Scalable Numerical Software on an O(100M) core environment
作成日時 01/04/2015–31/03/2018
Offer Organization: Japan Society for the Promotion of Science, System Name: Grants-in-Aid for Scientific Research, Category: Grant-in-Aid for Scientific Research (B), Fund Type: -, Overall Grant Amount: - (direct: 14100000, indirect: 4230000)
This research project aims to realize high performance numerical services investigated in the past based on new mathematical principles in the emerging computing system where tens of thousands to hundreds of millions of processing cores are installed. Giving two important themes, `Mixed-granularity numerical kernel' and `Asynchronous numerical algorithm,' we conducted; i) the research on the theory of asynchronous numerical algorithms. Also avoidance of communication and synchronization at a practical level, then CAHTR and a new method for the FDTD scheme were proposed. Furthermore, we have practiced; ii) promoting research on core numerical infrastructure technologies such as automatic tuning for scalable, lightweight code generation at super-many-core, and promoting innovative research leading to the next generation numerical calculation software.
その他
作成日時 01/04/2015–31/03/2018
Offer Organization: Japan Society for the Promotion of Science, System Name: Grants-in-Aid for Scientific Research, Category: Grant-in-Aid for Scientific Research (B), Fund Type: -, Overall Grant Amount: - (direct: 14000000, indirect: 4200000)
Autotuning is technology that aims to attain good performance under various conditions, by letting software controls its own parameter. In case multiple parameter exist, most of previous research chose either exhaustive search or heuristic pruning. In this research, we aim mathematically founded method using Bayesian statistics, which gives practically good and asymptotically optimal solutions.
In survey of previous works, we found that linear models and correlation models have such properties, and they can be combined. From description of such models, we create a software that generates a code that constructs performance model from a priori information and observations. Also we apply autotuning mathematical libraries to various computations, and confirms their effectiveness.