2026/03/11 更新

写真a

ヨコタ リオ
横田 理央
YOKOTA RIO
所属
総合研究院 スーパーコンピューティング研究センター 教授
職名
教授
外部リンク

学位

  • 博士(工学) ( 慶應義塾大学 )

研究キーワード

  • 数値解析

  • 深層学習

  • 高性能計算

  • GPU

研究分野

  • ものづくり技術(機械・電気電子・化学工学) / 流体工学

  • 情報通信 / 知能情報学

  • 情報通信 / 数理情報学

  • 情報通信 / 計算科学

  • 情報通信 / 高性能計算

学歴

  • 慶應義塾大学   理工学研究科   開放環境科学専攻 博士課程

    2005年4月 - 2009年3月

      詳細を見る

  • 慶應義塾大学   理工学研究科   開放環境科学専攻 修士課程

    2003年4月 - 2005年3月

      詳細を見る

    国名: 日本国

    researchmap

  • 慶應義塾大学   理工学部   機械工学科

    1997年4月 - 2003年3月

      詳細を見る

    国名: 日本国

    researchmap

経歴

  • 東京科学大学   総合研究院 スーパーコンピューティング研究センター   教授

    2024年10月 - 現在

      詳細を見る

  • 東京工業大学   学術国際情報センター   教授

    2023年1月 - 2024年9月

      詳細を見る

  • 東京工業大学   学術国際情報センター   准教授

    2015年4月 - 2022年12月

      詳細を見る

  • アブドゥラ王立科学技術大学   応用数学・情報工学科   研究員

    2011年9月 - 2015年3月

      詳細を見る

  • ボストン大学   機械工学科   ポスドク研究員

    2010年9月 - 2011年8月

      詳細を見る

  • ブリストル大学   数学科   ポスドク研究員

    2009年2月 - 2010年8月

      詳細を見る

▼全件表示

所属学協会

▼全件表示

委員歴

  • IEEE International Conference on Cluster Computing (IEEE CLUSTER 2025)   track co-chair  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • Platform for Advanced Scientific Computing (PASC 2025)   domain co-chair  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 39th IEEE International Parallel and Distributed Processing Symposium, (IPDPS 2025)   program committee, best OSS judge  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • International Conference on Parallel Architectures and Compilation Techniques (PACT 2025)   publicity chair  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2025)   proceedings vice-chair, program committee, AI4S committee  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • Conference on Neural Information Processing Systems (NeurIPS 2025)   reviewer  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • ISC High Performance (ISC 2025)   program committee  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • International Conference on Machine Learning (ICML 2025)   reviewer  

    2025年   

      詳細を見る

    団体区分:学協会

    researchmap

  • SIAM Conference on Linear Algebra (LA24), scientific committee  

    2024年   

      詳細を見る

  • The IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR 2024), reviewer  

    2024年   

      詳細を見る

  • 38th Conference on Neural Information Processing Systems (NeurIPS 2024), reviewer  

    2024年   

      詳細を見る

  • The 30th International Conference on Parallel Processing (Euro-Par 2024), program committee  

    2024年   

      詳細を見る

  • ACM International Symposium on High-Performance, Parallel and Distributed Computing (HPDC 2024), program committee  

    2024年   

      詳細を見る

  • SC Asia (SCA 2024), program committee  

    2024年   

      詳細を見る

  • Platform for Advanced Scientific Computing (PASC 2024), program committee  

    2024年   

      詳細を見る

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2024), poster, ML track, AI4S/TPC WS, LLMs for HPC BoF  

    2024年   

      詳細を見る

  • International Conference on Preconditioning Techniques for Scientific and Industrial Applications, (Precond 2024), program committee  

    2024年   

      詳細を見る

  • 38th IEEE International Parallel and Distributed Processing Symposium, (IPDPS 2024), track co-chair  

    2024年   

      詳細を見る

  • ISC High Performance (ISC 2024), posters chair  

    2024年   

      詳細を見る

  • The 12th International Conference on Learning Representations (ICLR 2024), reviewer  

    2024年   

      詳細を見る

  • The Eleventh International Conference on Learning Representations (ICLR 2023), reviewer  

    2023年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2023), posters, workshops, ML tech paper, panelist, best paper  

    2023年   

      詳細を見る

  • 37th Conference on Neural Information Processing Systems (NeurIPS 2023), reviewer  

    2023年   

      詳細を見る

  • 10th International Congress on Industrial and Applied Mathematics (ICIAM 2023), program committee  

    2023年   

      詳細を見る

  • Platform for Advanced Scientific Computing (PASC 2023), mini symposia and posters com- mittee, program committee  

    2023年   

      詳細を見る

  • The 23rd IEEE/ACM international Symposium on Cluster, Cloud and Internet Computing (CCGrid 2023), program committee  

    2023年   

      詳細を見る

  • The IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR 2023), re- viewer  

    2023年   

      詳細を見る

  • The 32nd International Conference on Parallel Architectures and Compilation Techniques (PACT 2023), publicity chair Asia  

    2023年   

      詳細を見る

  • IEEE International Conference on Cluster Computing (IEEE CLUSTER 2023), publicity chair  

    2023年   

      詳細を見る

  • Principles and Practice of Parallel Programming (PPoPP 2023), publicity chair  

    2023年   

      詳細を見る

  • 37th IEEE International Parallel and Distributed Processing Symposium, (IPDPS 2023), pro- gram committee  

    2023年   

      詳細を見る

  • ISC High Performance (ISC 2023), posters deputy chair  

    2023年   

      詳細を見る

  • SIAM Conference on Computational Science and Engineering (CSE23), poster judge  

    2023年   

      詳細を見る

  • IEEE Cluster (IEEE Cluster 2022), program committee  

    2022年   

      詳細を見る

    団体区分:学協会

    researchmap

  • International Conference on High Performance Computing ini Asia-Pacific Region (HPC Asia 2022), Applications track co-chair  

    2022年   

      詳細を見る

    団体区分:学協会

    researchmap

  • European Conference on Computer Vision (ECCV 2022), reviewer  

    2022年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2022), program committee, workshop chair  

    2022年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 49th International Conference on Parallel Processing (ICPP 2022), program committee  

    2022年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 36th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2022), program committee  

    2022年   

      詳細を見る

    団体区分:学協会

    researchmap

  • ISC High Performance (ISC 2022), poster committee  

    2022年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE Cluster (IEEE Cluster 2021), program committee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • ISC High Performance (ISC 2021), program committee, best paper committee, steering com- mittee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 35th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2021), program committee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • International Conference on High Performance Computing ini Asia-Pacific Region (HPC Asia 2021), organizing committee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • SIAM Computational Science and Engineering (SIAM CSE 2021), organizing committee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD 2021), program committee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 48th International Conference on Parallel Processing (ICPP 2021), program committee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2021), program committee  

    2021年   

      詳細を見る

    団体区分:学協会

    researchmap

  • SIAM Conference on Parallel Processing for Scientific Computing (SIAM PP 2020), program committee  

    2020年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2020), ML track Chair  

    2020年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 26th International Conference on Parallel Processing (Euro-Par 2020), program committee  

    2020年   

      詳細を見る

    団体区分:学協会

    researchmap

  • International Conference on High Performance Computing ini Asia-Pacific Region (HPC Asia 2020), program committee  

    2020年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 48th International Conference on Parallel Processing (ICPP 2019), Applications track chair, Workshop co-organizer  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • SIAM Conference on Computational Science and Engineering (SIAM CSE19), organizing com- mittee  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2019), program committee  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • ISC High Performance (ISC 2019), program committee  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE Cluster (IEEE Cluster 2019), program committee  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • Platform for Advanced Scientific Computing Conference (PASC 2019), program committee  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 3rd cross-disciplinary Workshop on Computing Systems, Infrastructures, and Program- ming (xSIG 2019), program committee  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE 13th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC 2019), ATMG Track Chair  

    2019年   

      詳細を見る

    団体区分:学協会

    researchmap

  • SIAM Conference on Parallel Processing for Scientific Computing (SIAM PP 2018), organizing committee  

    2018年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2018), proceedings chair  

    2018年   

      詳細を見る

    団体区分:学協会

    researchmap

  • ISC High Performance (ISC 2018), proceedings chair  

    2018年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE Cluster (IEEE Cluster 2018), program committee  

    2018年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 32nd IEEE International Parallel & Distributed Processing Symposium (IPDPS 2018), program committee  

    2018年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 18th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 2018), program committee  

    2018年   

      詳細を見る

    団体区分:学協会

    researchmap

  • SC Asia (SCA 2018), program committee  

    2018年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 23rd International Conference on Parallel Processing (Euro-Par 2017), program committee  

    2017年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2017), program committee  

    2017年   

      詳細を見る

    団体区分:学協会

    researchmap

  • Platform for Advanced Scientific Computing Conference (PASC 2017), program committee  

    2017年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 31st IEEE International Parallel & Distributed Processing Symposium (IPDPS 2017), program committee  

    2017年   

      詳細を見る

    団体区分:学協会

    researchmap

  • International Conference on Supercomputing (ICS 2017), program committee  

    2017年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE Cluster (IEEE Cluster 2017), program committee  

    2017年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • International Conference on Supercomputing (ICS 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • Platform for Advanced Scientific Computing Conference (PASC 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • ISC High Performance (ISC 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 30th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 23rd annual IEEE International Conference on High Performance Computing, Data, and Analytics (HiPC 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 28th International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE Cluster (IEEE Cluster 2016), program committee  

    2016年   

      詳細を見る

    団体区分:学協会

    researchmap

  • IEEE Cluster (IEEE Cluster 2015), program committee  

    2015年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2015), program committee  

    2015年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 22nd annual IEEE International Conference on High Performance Computing, Data, and Analytics (HiPC 2015), program committee  

    2015年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 15th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 2015), program committee  

    2015年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 14th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid 2014), program committee  

    2014年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The annual IEEE International Conference on High Performance Computing (HiPC 2014), program committee  

    2014年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 20th International Conference on Parallel Processing (Euro-Par 2014), program committee  

    2014年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2014), program committee  

    2014年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Meeting on High-Performance Computing for Computational Science (VEC- PAR 2014), program committee  

    2014年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2013), program committee  

    2013年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 28th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2013), program committee  

    2013年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 18th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP 2013), program committee  

    2013年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The annual IEEE International Conference on High Performance Computing (HiPC 2013), program committee  

    2013年   

      詳細を見る

    団体区分:学協会

    researchmap

  • 15th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (SYNASC 2013), program committee  

    2013年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The Third International Workshop on Frontier of GPU Computing, program committee  

    2012年   

      詳細を見る

    団体区分:学協会

    researchmap

  • The 12th International Symposium on Parallel and Distributed Computing (ISPDC 2012), program committee  

    2012年   

      詳細を見る

    団体区分:学協会

    researchmap

▼全件表示

論文

  • Tensor-Core-Optimized Strategies for BLR × Tall-Skinny Matrix Multiplication in BEM

    Akihiro Ida, Kazuya Goto, Rio Yokota, Tasuku Hiraishi, Toshihiro Hanawa, Takeshi Iwashita, Masatoshi Kawai, Satoshi Ohshima, Tetsuya Hoshino

    Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region   153 - 164   2026年1月

     詳細を見る

    掲載種別:研究論文(国際会議プロシーディングス)   出版者・発行元:ACM  

    DOI: 10.1145/3773656.3773678

    researchmap

  • A Study on the Performance and Usability of Managed Memory and Unified Memory for Accelerating Numerical Calculation Program

    Satoshi Ohshima, Akihiro Ida, Masatoshi Kawai, Takeshi Fukaya, Rio Yokota

    2025 IEEE 18th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC)   41 - 48   2025年12月

     詳細を見る

    掲載種別:研究論文(国際会議プロシーディングス)   出版者・発行元:IEEE  

    DOI: 10.1109/mcsoc67473.2025.00017

    researchmap

  • Variational Learning Finds Flatter Solutions at the Edge of Stability 査読 国際誌

    Rio Yokota

    Annual Conference on Neural Information Processing Systems (NeurIPS)   2025年12月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Masked Gated Linear Unit 査読

    Yukito Tajima, Nakamasa Inoue, Yusuke Sekikawa, Ikuro Sato, Rio Yokota

    Annual Conference on Neural Information Processing Systems (NeurIPS)   2025年12月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Leveraging High-Resource English Corpora for Cross-lingual Domain Adaptation in Low-Resource Japanese Medicine via Continued Pre-training 査読 国際誌

    Rio Yokota

    EMNLP Findings   2025年11月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Building Instruction-Tuning Datasets from Human-Written Instructions with Open-Weight Large Language Models 査読

    Youmi Ma, Sakae Mizuki, Kazuki Fujii, Taishi Nakamura, Masanari Ohi, Hinari Shimada, Taihei Shiotani, Koshiro Saito, Koki Maeda, Kakeru Hattori, Takumi Okamoto, Shigeki Ishida, Rio Yokota, Hiroya Takamura, Naoaki Okazaki

    Conference on Language Modeling (COLM)   2025年10月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Optimal Sparsity of Mixure-of-Experts Language Models for Reasoning Tasks 査読

    Taishi Nakamura, Satoki Ishikawa, Masaki Kawamura, Takumi Okamoto, Daisuke Nohara, Jun Suzuki, Rio Yokota

    ICML 2025 2nd AI for Math Workshop   2025年7月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Lion Cub: Minimizing Communication Overhead in Distributed Lion 査読

    Satoki Ishikawa, Tal Ben-Nun, Brian Van Essen, Rio Yokota, Nikoli Dryden

    ICML 2025 Workshop TTODLer-FM   2025年7月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Scaling Large-scale GNN Training to Thousands of Processors on CPU-based Supercomputers

    Chen Zhuang, Lingqi Zhang, Du Wu, Peng Chen, Jiajun Huang, Xin Liu, Rio Yokota, Nikoli Dryden, Toshio Endo, Satoshi Matsuoka, Mohamed Wahib

    Proceedings of the 39th ACM International Conference on Supercomputing   57 - 72   2025年6月

     詳細を見る

    掲載種別:研究論文(国際会議プロシーディングス)   出版者・発行元:ACM  

    DOI: 10.1145/3721145.3730422

    researchmap

  • Improving LoRA with Variational Learning.

    Bai Cong, Nico Daheim, Yuesong Shen, Rio Yokota, Mohammad Emtiyaz Khan, Thomas Möllenhoff

    CoRR   abs/2506.14280   2025年6月

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    DOI: 10.48550/arXiv.2506.14280

    researchmap

  • On the Interplay Between Precision, Rank, Admissibility, and Iterative Refinement for Hierarchical Low-Rank Matrix Solvers 査読

    Thomas Spendlhofer, Qianxiang Ma, Yasuhiro Matsumoto, Rio Yokota

    ISC High Performance   2025年6月

     詳細を見る

    記述言語:英語  

    researchmap

  • Rewriting Pre-Training Data Boosts LLM Performance in Math and Code.

    Kazuki Fujii, Yukito Tajima, Sakae Mizuki, Hinari Shimada, Taihei Shiotani, Koshiro Saito, Masanari Ohi, Masaki Kawamura, Taishi Nakamura, Takumi Okamoto, Shigeki Ishida, Kakeru Hattori, Youmi Ma, Hiroya Takamura, Rio Yokota, Naoaki Okazaki

    CoRR   abs/2505.02881   2025年5月

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    DOI: 10.48550/arXiv.2505.02881

    researchmap

  • Quantum Turbulence Coupled with Externally Driven Normal-Fluid Turbulence in Counterflow of Superfluid 4He 査読

    Satoshi Yui, Hiromichi Kobayashi, Makoto Tsubota, Rio Yokota

    Journal of the Physical Society of Japan   94 ( 4 )   2025年4月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)   出版者・発行元:Physical Society of Japan  

    DOI: 10.7566/jpsj.94.043601

    researchmap

  • Local Loss Optimization in the Infinite Width: Stable Parameterization of Predictive Coding Networks and Target Propagation 査読

    Satoki Ishikawa, Ryo Karakida, Rio Yokota

    The 13th International Conference on Learning Representations (ICLR)   2025年4月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Drop Upcycling: Training Sparse Mixture of Experts with Partial Re-initialization 査読

    Taishi Nakamura, Takuya Akiba, Kazuki Fujii, Yusuke Oda, Rio Yokota, Jun Suzuki

    The 13th International Conference on Learning Representations (ICLR)   2025年4月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • SuperGCN: General and Scalable Framework for GCN Training on CPU-powered Supercomputers 査読

    Chen Zhuang, Peng Chen, Xin Liu, Nikoli Dryden, Rio Yokota, Toshio Endo, Satoshi Matsuoka, Mohamed Wahib

    ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP)   2025年3月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • 新聞記事と合成データを用いた日本語LLMの継続事前学習

    服部翔, 水木栄, 藤井一喜, 中村泰士, 塩谷泰平, 植木快, 新妻巧朗, 田森秀明, Youmi Ma, 前田航希, 大井聖也, 齋藤幸史郎, 岡本拓己, 石田茂樹, 横田理央, 高村大也, 岡崎直観

    言語処理学会第31回年次大会   2025年3月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 模倣学習による大規模言語モデルの指示チューニング

    Youmi Ma, 水木栄, 藤井一喜, 中村泰士, 大井聖也, 島田比奈理, 塩谷泰平, 齋藤幸史郎, 前田航希, 服部翔, 岡本拓己, 石田茂樹, 横田理央, 高村大也, 岡崎直観

    言語処理学会第31回年次大会   2025年3月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Aurora-M: The First Open Source Multilingual Language Model Red-teamed according to the U.S. Executive Order 査読

    Taishi Nakamura, Mayank Mishra, Simone Tedeschi, Yekun Chai, Jason T Stillerman, Felix Friedrich, Prateek Yadav, Tanmay Laud, Vu Minh Chien, Terry Yue Zhuo, et al.

    The 31st International Conference on Computational Linguistics (COLING), Industry Track   2025年1月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • A General and Scalable GCN Training Framework on CPU Supercomputers.

    Chen Zhuang, Peng Chen 0035, Xin Liu 0020, Rio Yokota, Nikoli Dryden, Lingqi Zhang 0001, Toshio Endo, Satoshi Matsuoka, Mohamed Wahib

    PPoPP   566 - 568   2025年

     詳細を見る

    掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1145/3710848.3710860

    researchmap

    その他リンク: https://dblp.uni-trier.de/db/conf/ppopp/ppopp2025.html#ZhuangCLYD0EMW25

  • On the Relationship Between Double Descent of CNNs and Shape/Texture Bias Under Learning Process

    Shun Iwase, Shuya Takahashi, Nakamasa Inoue, Rio Yokota, Ryo Nakamura, Hirokatsu Kataoka, Eisaku Maeda

    2025年

  • Rethinking Image Super-Resolution from Training Data Perspectives 査読

    Go Ohtani, Ryu Tadokoro, Ryosuke Yamada, Yuki M. Asano, Iro Laina, Christian Rupprecht, Nakamasa Inoue, Rio Yokota, Hirokatsu Kataoka, Yoshimitsu Aoki

    European Conference on Computer Vision (ECCV)   2025年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-031-72643-9_2

    researchmap

  • Scaling Backwards: Minimal Synthetic Pre-Training? 査読

    Ryo Nakamura, Ryu Tadokoro, Ryosuke Yamada, Yuki M. Asano, Iro Laina, Christian Rupprecht, Nakamasa Inoue, Rio Yokota, Hirokatsu Kataoka

    European Conference on Computer Vision (ECCV)   2025年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-031-72633-0_9

    researchmap

  • A Framework for Seamless Integration and Efficient Continual Pre-Training of Large Language Models 査読

    Kazuki Fujii, Taishi Nakamura, Rio Yokota

    SC’24 TPC Workshop   2024年11月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Continual Pre-Training for Cross-Lingual LLM Adaptation: Enhancing Japanese Language Capabilities 査読

    Kazuki Fujii, Taishi Nakamura, Mengsay Loem, Hiroki Iida, Masanari Ohi, Kakeru Hattori, Shota Hirai, Sakae Mizuki, Rio Yokota, Naoaki Okazaki

    Conference on Language Modeling COLM   2024年10月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Building a Large Japanese Web Corpus for Large Language Models 査読

    Naoaki Okazaki, Kakeru Hattori, Shota Hirai, Hiroki Iida, Masanari Ohi, Kazuki Fujii, Taishi Nakamura, Mengsay Loem, Rio Yokota, Sakae Mizuki

    Conference on Language Modeling COLM   2024年10月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Formula-Supervised Visual-Geometric Pre-training 査読

    Ryosuke Yamada, Kensho Hara, Hirokatsu Kataoka, Koshi Makihara, Nakamasa Inoue, Rio Yokota, Yutaka Satoh

    European Conference on Computer Vision (ECCV)   2024年9月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-031-72670-5_4

    researchmap

  • 画像超解像における学習データ構築の再考 査読

    大谷 豪, 田所 龍, 山田 亮佑, Yuki M. asano, Iro Laina, Chistian Repprech, 井上 中順, 横田 理央, 片岡 裕雄, 青木 義満

    第27回 画像の認識・理解シンポジウム (MIRU)   2024年8月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Scaling Backwards: Minimal Synthetic Pre-training? 査読

    田所 龍, 中村 凌, 山田 亮佑, Yuki M. Asano, Iro Laina, Chistian Repprech, 井上 中順, 横田 理央, 片岡 裕雄

    第27回 画像の認識・理解シンポジウム (MIRU)   2024年8月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 単眼カメラを用いたリアルタイムな3次元マップの変化検出を目的とした密なバンドル調整 査読

    大川快, 櫻田健, 横田理央

    第27回 画像の認識・理解シンポジウム (MIRU)   2024年8月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • An inherently parallel ℋ2-ULV factorization for solving dense linear systems on GPUs 査読

    Qianxiang Ma, Rio Yokota

    The International Journal of High Performance Computing Applications   2024年7月

     詳細を見る

    担当区分:最終著者   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1177/10943420241242021

    researchmap

  • Variational Learning is Effective for Large Deep Networks 査読

    Yuesong Shen, Nico Daheim, Gian Maria Marconi, Peter Nickl, Bai Cong, Bazan Clemen, Emile Marcel Raoul, Rio Yokota, Iryna Gurevych, Daniel Cremers, Mohammad, Emtiyaz Khan, Thomas Möllenhoff

    The 41st International Conference on Machine Learning (ICML)   2024年7月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Interaction between quantum turbulence and normal-fluid turbulence in superfluid helium 査読

    Hiromichi Kobayashi, Satoshi Yui, Makoto Tsubota, Tomokazu Saito, Rio Yokota

    Thirteenth International Symposium on Turbulence and Shear Flow Phenomena (TSFP13)   2024年6月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • SIFTer: Self-improving Synthetic Datasets for Pre-training Classification Models 査読

    Ryo Hayamizu, Shota Nakamura, Sora Takashima, Hirokatsu Kataoka, Ikuro Sato, Nakamasa Inoue, Rio Yokota

    CVPR SynData Workshop   2024年6月

     詳細を見る

    担当区分:最終著者, 責任著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • One-Shot NASによるBERTのモデル圧縮

    岡本拓己, 横田理央

    人工知能学会全国大会 (第38回)   2024年5月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • When Does Second-Order Optimization Speed Up Training? 査読

    Satoki Ishikawa, Rio Yokota

    The 12th International Conference on Learning Representations (ICLR), Tiny paper   2024年5月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • DGEMM on integer matrix multiplication unit 査読

    Hiroyuki Ootomo, Katsuhisa Ozaki, Rio Yokota

    The International Journal of High Performance Computing Applications   38 ( 4 )   297 - 313   2024年3月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(学術雑誌)   出版者・発行元:SAGE Publications  

    Deep learning hardware achieves high throughput and low power consumption by reducing computing precision and specializing in matrix multiplication. For machine learning inference, fixed-point value computation is commonplace, where the input and output values and the model parameters are quantized. Thus, many processors are now equipped with fast integer matrix multiplication units (IMMU). It is of significant interest to find a way to harness these IMMUs to improve the performance of HPC applications while maintaining accuracy. We focus on computing double-precision equivalent matrix multiplication using the Ozaki scheme, which computes a high-precision matrix multiplication by using lower-precision computing units, and show the advantages and disadvantages of using IMMU. The experiment using integer Tensor Cores shows that we can compute double-precision matrix multiplication faster than cuBLAS and an existing Ozaki scheme implementation on FP16 Tensor Cores on NVIDIA consumer GPUs. Furthermore, we demonstrate accelerating a quantum circuit simulation by up to 4.85 while maintaining the FP64 accuracy.

    DOI: 10.1177/10943420241239588

    researchmap

    その他リンク: https://journals.sagepub.com/doi/full-xml/10.1177/10943420241239588

  • 継続事前学習による日本語に強い大規模言語モデルの構築

    藤井一喜, 中村泰士, Mengsay Loem, 飯田大貴, 大井聖也, 服部翔, 平井翔太, 水木栄, 横田理央, 岡崎直観

    言語処理学会第30回年次大会   2024年3月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 大規模言語モデルの分散並列学習

    藤井一喜, 横田理央

    情報処理学会全国大会   2024年3月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 大規模言語モデルの構造探索

    岡本拓己, 横田理央

    情報処理学会全国大会   2024年3月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Swallowコーパス: 日本語大規模ウェブコーパス

    岡崎直観, 服部翔, 平井翔太, 飯田大貴, 大井聖也, 藤井一喜, 中村泰士, Mengsay Loem, 横田理央, 水木栄

    言語処理学会第30回年次大会   2024年3月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 大規模言語モデルの日本語能力の効率的な強化: 継続事前学習における語彙拡張と対訳コーパスの活用

    水木栄, 飯田大貴, 藤井一喜, 中村泰士, Mengsay Loem, 大井聖也, 服部翔, 平井翔太, 横田理央, 岡崎直観

    言語処理学会第30回年次大会   2024年3月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 継続学習を用いた効率の良いマルチリンガル・マルチエキスパートモデルの開発

    中村泰士, 横田理央

    情報処理学会全国大会   2024年3月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Visual SLAM を目的とした深度の一貫性を考慮した密なバンドル調整

    大川 快, 櫻田 健, 横田 理央

    CVIM 2024年1月研究会   2024年1月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Natural Gradient Primal-Dual Method for Decentralized Learning. 査読

    Kenta Niwa, Hiro Ishii, Hiroshi Sawada, Akinori Fujino, Noboru Harada, Rio Yokota

    IEEE Trans. Signal Inf. Process. over Networks   10   417 - 433   2024年

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1109/TSIPN.2024.3388948

    researchmap

  • Why We Build Local Large Language Models: An Observational Analysis from 35 Japanese and Multilingual LLMs.

    Koshiro Saito, Sakae Mizuki, Masanari Ohi, Taishi Nakamura, Taihei Shiotani, Koki Maeda, Youmi Ma, Kakeru Hattori, Kazuki Fujii, Takumi Okamoto, Shigeki Ishida, Hiroya Takamura, Rio Yokota, Naoaki Okazaki

    CoRR   abs/2412.14471   2024年

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    DOI: 10.48550/arXiv.2412.14471

    researchmap

  • Variational Low-Rank Adaptation Using IVON.

    Bai Cong, Nico Daheim, Yuesong Shen, Daniel Cremers, Rio Yokota, Mohammad Emtiyaz Khan, Thomas Möllenhoff

    CoRR   abs/2411.04421   2024年

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    DOI: 10.48550/arXiv.2411.04421

    researchmap

  • 超解像事前学習における核心的要素の解明

    大谷 豪, 田所 龍, 片岡 裕雄, 井上 中順, 横田 理央, 青木 義満

    ViEW2023   2023年12月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Improving Continual Learning by Accurate Gradient Reconstructions of the Past 査読

    Erik Daxberger, Siddharth Swaroop, Kazuki Osawa, Rio Yokota, Richard E. Turner, Jose Miguel Hernandez-Lobato, Mohammad, Emtiyaz Khan

    Transactions on Machine Learning Research   2023年11月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    researchmap

  • Two-way coupled simulation of quantum turbulence and normal-fluid turbulence in superfluid helium-4

    Hiromichi Kobayashi, Satoshi Yui, Makoto Tsubota, Tomokazu Saito, Rio Yokota

    APS DFD 76th Annual Meeting of the Division of Fluid Dynamics   2023年11月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • SegRCDB: Semantic Segmentation via Formula-Driven Supervised Learning 査読

    Risa Shinoda, Ryo Hayamizu, Kodai Nakashima, Nakamasa Inoue, Rio Yokota, Hirokatsu Kataoka

    IEEE/CVF International Conference on Computer Vision (ICCV)   2023年10月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Pre-training Vision Transformers with Very Limited Synthesized Images 査読

    Ryo Nakamura, Sora Takashima, Edgar Josafat Martinez-Noriega, Nakamasa Inoue, Hirokatsu Kataoka, Rio Yokota

    IEEE/CVF International Conference on Computer Vision (ICCV)   2023年10月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Cache Optimization and Performance Modeling of Batched, Small, and Rectangular Matrix Multiplication on Intel, AMD, and Fujitsu Processors 査読

    Sameer Deshmukh, Rio Yokota, George Bosilca

    ACM Transactions on Mathematical Software   49 ( 3 )   1 - 29   2023年9月

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    researchmap

  • Two-phase flow of quantum turbulence and normal-fluid turbulence in superfluid helium-4

    Hiromichi Kobayashi, Satoshi Yui, Makoto Tsubota, Rio Yokota

    ERCOFTAC Symposium on Engineering Turbulence Modeling and Measurements (ETMM)   2023年9月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • 超流動ヘリウム4の量子乱流:常流体乱流による渦糸バンドルの形成

    湯井悟志, 小林宏充, 坪田誠, 齋藤智和, 横田理央

    日本物理学会 第78回年次大会   2023年9月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Vortex-filament bundle induced by normal-fluid turbulence in turbulent superfluid helium-4

    Hiromichi Kobayashi, Satoshi Yui, Makoto Tsubota, Tomokazu Saito, Rio Yokota

    International Symposium on Quantum Fluids and Solids (QFS)   2023年8月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Visual Atoms: Pre-training Vision Transformers with Sinusoidal Waves 査読

    Sora Takashima, Ryoh Hayamizu, Nakamasa Inoue, Hirokatsu Kataoka, Rio Yokota

    IEEE/CVF Conference on Computer Vision and Pattern Recognition   2023年6月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • ラージバッチ学習における汎化性能の低下を抑制する正則化手法 査読

    中村 祥大, 横田 理央

    人工知能学会全国大会   2023年6月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Quantum Circuit Simulation by SGEMM Emulation on Tensor Cores and Automatic Precision Selection 査読

    Hiroyuki Ootomo, Hidetaka Manabe, Kenji Harada, Rio Yokota

    ISC High Performance   2023年5月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-031-32041-5_14

    researchmap

  • 深層学習における勾配の前処理法に関する検討

    石川 智貴, 横田 理央

    情報処理学会全国大会   2023年3月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 量子渦計算の高速多重極展開法を用いた高速化

    齋藤 智和, 横田 理央

    情報処理学会全国大会   2023年3月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • GPUとA64FXにおけるTransformerの性能比較

    中村 秋海, 横田 理央

    第188回HPC研究発表会   2023年3月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • 敵対的距離学習モジュールを用いた特徴変動に頑健な画像認識のための対照学習

    杉山 佳史, 片岡 裕雄, 横田 理央, 井上 中順

    パターン認識・メディア理解研究会 (PRMU)   2023年3月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • ニュートンフラクタル画像による事前学習効果

    近江 俊樹, 中村 凌, 片岡 裕雄, 井上 中順, 横田 理央

    情報処理学会全国大会   2023年3月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Reducing Shared Memory Footprint to Leverage High Throughput on Tensor Cores and its Flexible API Extension Library 査読

    Hiroyuki Ootomo, Rio Yokota

    HPC Asia   2023年2月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • 蒸留画像による事前学習効果についての検討 査読

    田所龍, 片岡裕雄, 川上玲, 横田理央, 井上中順

    ビジョン技術の実利用ワークショップ   2022年12月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • ASDL: A Unified Interface for Gradient Preconditioning in PyTorch 査読

    Kazuki Osawa, Satoki Ishikawa, Rio Yokota, Shigang Li, Torsten Hoefler

    NeurIPS Workshop Order up! The Benefits of Higher-Order Optimization in Machine Learning   2022年12月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Empirical Study on Optimizer Selection for Out-of-Distribution Generalization 査読

    Hiroki Naganuma, Kartik Ahuja, Ioannis Mitliagkas, Shiro Takagi, Tetsuya Motokawa, Rio Yokota, Kohta Ishikawa, Ikuro Sato

    NeurIPS Workshop Distshift   abs/2211.08583   2022年12月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.48550/arXiv.2211.08583

    researchmap

  • QR Factorization of Block Low-Rank Matrices on Multi-Instance GPU 査読

    Satoshi Ohshima, Akihiro Ida, Rio Yokota, Ichitaro Yamazaki

    The 23rd International Conference on Parallel and Distributed Computing, Applications and Technologies (PDCAT’22)   2022年12月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Tensorコアを用いた単精度行列積エミュレーションのアプリケーションでの評価

    大友広幸, 横田 理央

    第185回ハイパフォーマンスコンピューティング研究発表会   2022年12月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Scalable Linear Time Dense Direct Solver for 3-D Problems Without Trailing Sub-Matrix Dependencies 査読

    Qianxiang Ma, Sameer Deshmukh, Rio Yokota

    The International Conference for High Performance Computing, Networking, Storage, and Analysis (SC22)   2022年11月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • 対称ブロック低ランク行列の精度保証付き固有値問題解法

    伊田 明弘, 荻田 武史, 横田 理央

    日本応用数理学会2022年度年会   2022   2022年9月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    J-GLOBAL

    researchmap

  • 走行動画の大規模自己教師あり学習の検討と計画 査読

    高橋那弥, 八嶋晋吾, 石川康太, 佐藤育郎, 横田理央

    第25回 画像の認識・理解シンポジウム   2022年7月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Replacing Labeled Real-image Datasets with Auto-generated Contours 査読

    Hirokatsu Kataoka, Ryo Hayamizu, Ryosuke Yamada, Kodai Nakashima, Sora Takashima, Xinyu Zhang, Edgar Josafat Martinez-Noriega, Nakamasa Inoue, Rio Yokota

    IEEE/CVF Conference on Computer Vision and Pattern Recognition   2022年6月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • TensorCoreを用いた精度補正単精度行列積

    大友 広幸, 横田 理央

    第180回ハイパフォーマンスコンピューティング研究発表会   2022年6月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • OPIRL: Sample Efficient Off-Policy Inverse Reinforcement Learning via Distribution Matching 査読

    Hana Hoshino, Kei Ota, Asako Kanezaki, Rio Yokota

    Proceedings of IEEE International Conference on Robotics and Automation   2022年5月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(学術雑誌)  

    researchmap

  • Parallel QR Factorization of Block Low-Rank Matrices 査読

    Muhammad Ridwan Apriansyah, Rio Yokota

    ACM Transactions on Mathematical Software   2022年5月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1145/3538647

    researchmap

  • Recovering Single Precision Accuracy from Tensor Cores While Surpassing the FP32 Theoretical Peak Performance 査読

    Hiroyuki Ootomo, Rio Yokota

    The International Journal of High Performance Computing Application   2022年2月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1177/10943420221090256

    researchmap

  • RePOSE: Real-Time Iterative Rendering and Refinement for 6D Object Pose Estimation 査読

    Shun Iwase, Xingyu Liu, Rawal Khirodkar, Rio Yokota, Kris M. Kitani

    Proceedings of the International Conference on Computer Vision   2021年10月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    researchmap

  • Self-supervised Continual Pretraining for Class Incremental Image Classification 査読

    Hikaru Nakata, Nakamasa Inoue, Rio Yokota

    Proceedings CVPR CLVISION Workshop (Findings)   2021年6月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • ExaFMM: a high-performance fast multipole method library with C++ and Python interfaces 査読

    Tingyu Wang, Rio Yokota, Lorena A. Barba

    The Journal of Open Source Software   6 ( 61 )   3145   2021年5月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.21105/joss.03145

    researchmap

  • Rich Information is Affordable: A Systematic Performance Analysis of Second-order Optimization Using K-FAC 査読

    Yuichiro Ueno, Kazuki Osawa, Yohei Tsuji, Akira Naruse, Rio Yokota

    Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining   2145 - 2153   2020年8月

     詳細を見る

    掲載種別:研究論文(国際会議プロシーディングス)   出版者・発行元:ACM  

    DOI: 10.1145/3394486.3403265

    Scopus

    researchmap

    その他リンク: https://dblp.uni-trier.de/conf/kdd/2020

  • Scalable and Practical Natural Gradient for Large-Scale Deep Learning 査読

    Kazuki Osawa, Yohei Tsuji, Yuichiro Ueno, Akira Naruse, Chuan-Sheng Foo, Rio Yokota

    IEEE Transactions on Pattern Analysis and Machine Intelligence   1 - 1   2020年6月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)   出版者・発行元:Institute of Electrical and Electronics Engineers ({IEEE})  

    DOI: 10.1109/TPAMI.2020.3004354

    researchmap

  • Effect of Mixed Precision Computing on H-Matrix Vector Multiplication in BEM Analysis 査読

    Rise Ooi, Takeshi Iwashita, Takeshi Fukaya, Akihiro Ida, Rio Yokota

    ACM International Conference Proceeding Series   92 - 101   2020年1月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1145/3368474.3368479

    Web of Science

    Scopus

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1911.00093v1

  • ニューラルネットワーク学習におけるLARSによるバッチサイズ拡張効果と反復回数削減効果の検証

    長沼 大樹, 井手 達朗, 横田 理央

    人工知能学会全国大会論文集   JSAI2020   4Rin169 - 4Rin169   2020年

     詳細を見る

    記述言語:日本語   出版者・発行元:一般社団法人 人工知能学会  

    深層学習では極めて冗長な数のパラメータを持つ深層ニューラルネットワーク(DNN)を膨大な学習データを用いて学習することで他の機械学習手法を圧倒する高い性能を発揮している。一方で、DNNの学習には膨大な計算時間がかかるため、大規模並列化によって学習時間を短縮するのが喫緊の課題である。高速化を目的とした大きなバッチサイズでの深層学習であるラージバッチ学習では、バッチサイズを大きくすることで学習に必要な反復回数を削減し高速化を達成するが、近年の研究で巨大なバッチサイズの領域では反復回数削減の効果に限界があることが示されている。本研究では、ラージバッチ学習において広く用いられる学習率調整法であるLARSを適用した場合に、フルバッチ規模までバッチサイズを増加させた範囲での実験を行い,バッチサイズと必要となる反復回数の関係を検証した。検証の結果、LARSの必要反復数の削減効果が他の最適化手法に比べて高いことが明らかとなったほか、フルバッチでの学習においても他の最適化手法に比べて汎化性能の劣化を防ぐことを実験的に明らかにした。

    DOI: 10.11517/pjsai.jsai2020.0_4rin169

    CiNii Research

    researchmap

  • Regularizing the fast multipole method for use in molecular simulation 査読

    D. S. Shamshirgar, R. Yokota, A. K. Tornberg, B. Hess

    Journal of Chemical Physics   151 ( 23 )   2019年12月

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    DOI: 10.1063/1.5122859

    Scopus

    PubMed

    researchmap

  • Practical deep learning with Bayesian principles 査読

    Kazuki Osawa, Siddharth Swaroop, Anirudh Jain, Runa Eschenhagen, Richard E. Turner, Rio Yokota, Mohammad Emtiyaz Khan

    Advances in Neural Information Processing Systems   32   4289 - 4301   2019年12月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    Web of Science

    Scopus

    researchmap

    その他リンク: https://dblp.uni-trier.de/conf/nips/2019

  • QR factorization of block low-rank matrices with weak admissibility condition 査読

    Akihiro Ida, Hiroshi Nakashima, Tasuku Hiraishi, Ichitaro Yamazaki, Rio Yokota, Takeshi Iwashita

    Journal of Information Processing   27   831 - 839   2019年11月

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    DOI: 10.2197/ipsjjip.27.831

    Scopus

    researchmap

  • Optimization of Numerous Small Dense-Matrix–Vector Multiplications in H-matrix Arithmetic on GPU 査読

    9 - 16   2019年10月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/MCSoC.2019.00009

    Web of Science

    Scopus

    researchmap

    その他リンク: https://dblp.uni-trier.de/db/conf/mcsoc/mcsoc2019.html#OhshimaYIY19

  • Distributed-memory lattice H-matrix factorization 査読

    Ichitaro Yamazaki, Akihiro Ida, Rio Yokota, Jack Dongarra

    International Journal of High Performance Computing Applications   33 ( 5 )   1046 - 1063   2019年9月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1177/1094342019861139

    Web of Science

    Scopus

    researchmap

  • Tensorコアを用いたTSQR 査読

    大友 広幸, 横田 理央

    日本応用数理学会年会   2019年9月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Performance optimizations and analysis of distributed deep learning with approximated second-order optimization method 査読

    Yohei Tsuji, Kazuki Osawa, Yuichiro Ueno, Akira Naruse, Rio Yokota, Satoshi Matsuoka

    ACM International Conference Proceeding Series   21 - 8   2019年8月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1145/3339186.3339202

    Web of Science

    Scopus

    researchmap

    その他リンク: https://dblp.uni-trier.de/db/conf/icppw/icppw2019.html#TsujiOUNYM19

  • Extreme scale FMM-accelerated boundary integral equation solver for wave scattering 査読

    Mustafa Abduljabbar, Mohammed Al Farhan, Noha Al-Harthi, Rui Chen, Rio Yokota, Hakan Bagci, David Keyes

    SIAM Journal on Scientific Computing   41 ( 3 )   C245 - C268   2019年6月

     詳細を見る

    掲載種別:研究論文(学術雑誌)  

    DOI: 10.1137/18M1173599

    Scopus

    researchmap

  • Large-scale distributed second-order optimization using kronecker-factored approximate curvature for deep convolutional neural networks 査読

    Kazuki Osawa, Yohei Tsuji, Yuichiro Ueno, Akira Naruse, Rio Yokota, Satoshi Matsuoka

    Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition   2019-June   12351 - 12359   2019年6月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/CVPR.2019.01264

    Web of Science

    Scopus

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1811.12019v5

  • Exhaustive study of hierarchical allreduce patterns for large messages between GPUs 査読

    Yuichiro Ueno, Rio Yokota

    Proceedings - 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2019   430 - 439   2019年5月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/CCGRID.2019.00057

    Web of Science

    Scopus

    researchmap

    その他リンク: https://dblp.uni-trier.de/db/conf/ccgrid/ccgrid2019.html#UenoY19

  • A Performance Improvement Approach for Second-Order Optimization in Large Mini-batch Training. 査読

    Proceedings - 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2019   696 - 703   2019年5月

     詳細を見る

    掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/CCGRID.2019.00092

    researchmap

    その他リンク: https://dblp.uni-trier.de/db/conf/ccgrid/ccgrid2019.html#NaganumaY19

  • Fisher情報行列の解析に基づく大規模深層学習のための二次最適化手法

    大沢 和樹, 横田 理央, Chuan-Sheng Foo, Vijay Chandrasekhar

    第81回情報処理学会全国大会講演論文集   2019 ( 1 )   45 - 46   2019年2月

     詳細を見る

    記述言語:日本語  

    画像データセットImageNetを始めとする巨大データセットを用いる大規模深層学習においては,膨大な学習時間が最適なパラメータ探索の障害となっている.学習時間の短縮を目的とした既存研究では,コスト関数の最小化に単純な一次最適化手法が用いられ,計算機の性能に頼った高速化手法が提案されてきた.一方で,自然勾配法は深層学習における効率的な二次最適化手法として知られているが,パラメータ数に依存するFisher情報行列の計算がボトルネックとなり,応用は限られていた.本研究では,これまで明らかにされてこなかった大規模深層学習におけるFisher情報行列の解析に基づき,より効率的な二次最適化手法を提案する.

    CiNii Books

    researchmap

  • Effect of Mixed Precision Computing on H-Matrix Vector Multiplication in BEM Analysis. 査読

    Rise Ooi, Takeshi Iwashita, Takeshi Fukaya, Akihiro Ida, Rio Yokota

    CoRR   abs/1911.00093   2019年

     詳細を見る

  • Highly productive, high-performance application frameworks for Post-Petascale computing 査読

    Naoya Maruyama, Takayuki Aoki, Kenjiro Taura, Rio Yokota, Mohamed Wahib, Motohiko Matsuda, Keisuke Fukuda, Takashi Shimokawabe, Naoyuki Onodera, Michel Müller, Shintaro Iwasaki

    Advanced Software Technologies for Post-Peta Scale Computing: The Japanese Post-Peta CREST Research Project   77 - 98   2018年12月

     詳細を見る

    記述言語:英語   掲載種別:論文集(書籍)内論文  

    DOI: 10.1007/978-981-13-1924-2_5

    Scopus

    researchmap

  • 自然勾配近似法を用いた大規模並列深層学習におけるハイパーパラメータ最適化 査読

    長沼大樹, 岩瀬 駿, 郭 林昇, 中田 光, 横田 理央

    第17回情報科学技術フォーラム   2018年9月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    researchmap

  • Performance of Hierarchical-matrix BiCGStab Solver on GPU Clusters 査読

    Ichitaro Yamazaki, Ahmad Abdelfattah, Akihiro Ida, Satoshi Ohshima, Stanimire Tomov, Rio Yokota, Jack Dongarra

    Proceedings - 2018 IEEE 32nd International Parallel and Distributed Processing Symposium, IPDPS 2018   930 - 939   2018年8月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/IPDPS.2018.00102

    Web of Science

    Scopus

    researchmap

    その他リンク: https://dblp.uni-trier.de/db/conf/ipps/ipdps2018.html#YamazakiAIOTYD18

  • Fast multipole preconditioners for sparse matrices arising from elliptic equations 査読

    Huda Ibeid, Rio Yokota, Jennifer Pestana, David Keyes

    Computing and Visualization in Science   18 ( 6 )   213 - 229   2018年3月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)   出版者・発行元:Springer Verlag  

    DOI: 10.1007/s00791-017-0287-5

    Scopus

    arXiv

    researchmap

  • Accelerating Convolutional Neural Networks Using Low Precision Arithmetic 査読

    HPC Asia   2018年1月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Optimization of hierarchical matrix computation on GPU 査読

    Satoshi Ohshima, Ichitaro Yamazaki, Akihiro Ida, Rio Yokota

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   10776 LNCS   274 - 292   2018年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-319-69953-0_16

    Web of Science

    Scopus

    researchmap

    その他リンク: https://dblp.uni-trier.de/db/conf/scfa/scfa2018.html#OhshimaYIY18

  • Fast Multipole Preconditioners for Sparse Matrices Arising from Elliptic Equations 査読

    Huda Ibeid, Rio Yokota, Jennifer Pestana, David Keyes

    Computing and Visualization in Science   Vol. 18 ( No. 6 )   pp. 213 - 229   2017年11月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    researchmap

  • Accelerating matrix multiplication in deep learning by using low-rank approximation 査読

    Kazuki Osawa, Akira Sekiya, Hiroki Naganuma, Rio Yokota

    Proceedings - 2017 International Conference on High Performance Computing and Simulation, HPCS 2017   186 - 192   2017年9月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/HPCS.2017.37

    Web of Science

    Scopus

    researchmap

  • Communication Reducing Algorithms for Distributed Heirarchical N-Body Methods 査読

    32nd International Conference, ISC High Performance   2017年6月

     詳細を見る

  • Communication Reducing Algorithms for Distributed Hierarchical N-Body Problems with Boundary Distributions 査読

    Mustafa Abduljabbar, George Markomanolis, Huda Ibeid, Rio Yokota, David Keyes

    Lecture Notes in Computer Science   10266   79 - 96   2017年2月

     詳細を見る

    記述言語:英語   出版者・発行元:Springer Verlag  

    DOI: 10.1007/978-3-319-58667-0_5

    Scopus

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1702.05459v1

  • Compute-Memory Tradeoff in Hierarchical Low-Rank Approximation Methods 査読

    SIAM Conference on Computational Science and Engineering   2017年2月

     詳細を見る

  • Fast Multipole Method as a Matrix-Free Hierarchical Low-Rank Approximation 査読

    Rio Yokota, Huda Ibeid, David Keyes

    EIGENVALUE PROBLEMS: ALGORITHMS, SOFTWARE AND APPLICATIONS IN PETASCALE COMPUTING (EPASA 2015)   117   267 - 286   2017年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-319-62426-6_17

    Web of Science

    Scopus

    researchmap

  • Performance evaluation of computation and communication kernels of the fast multipole method on intel manycore architecture 査読

    Mustafa Abduljabbar, Mohammed Al Farhan, Rio Yokota, David Keyes

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   10417   553 - 564   2017年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)   出版者・発行元:Springer Verlag  

    DOI: 10.1007/978-3-319-64203-1_40

    Scopus

    researchmap

  • Evaluating the compression efficiency of the filters in convolutional neural networks 査読

    Kazuki Osawa, Rio Yokota

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   10614 LNCS   459 - 466   2017年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-319-68612-7_52

    Web of Science

    Scopus

    researchmap

  • Tapas: An Implicitly Parallel ProgrammingFramework For Hierarchical N-body Algorithms 査読

    福田 圭祐, Fukuda Keisuke, Maruyama Naoya, 横田 理央, Yokota Rio, Taura Kenjiro, 松岡 聡, MATSUOKA SATOSHI

    The 22nd IEEE International Conference on Parallel And Distributed Systems   1100 - 1109   2016年12月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/ICPADS.2016.143

    Web of Science

    researchmap

  • Fast Multipole Preconditioners for Sparse Matrices Arising from Elliptic Equations 査読

    IBEID Huda, YOKOTA Rio, PESTANA Jennifer, KEYES David

    Computing and Visualization in Science   2016年12月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    researchmap

  • FMM と H^2(HSS) 行列のトレードオフについて 招待 査読

    横田 理央, Yokota Rio

    計算工学   21 ( 4 )   6 - 8   2016年10月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(学術雑誌)  

    researchmap

  • 大規模境界要素法解析における分散並列 FMM の通信最適化 招待 査読

    横田 理央, Yokota Rio

    シミュレーション   35 ( 3 )   2016年9月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(学術雑誌)   出版者・発行元:日本シミュレーション学会  

    researchmap

  • A Performance Model for the Communication in Fast Multipole Methods on HPC Platforms 査読

    Huda Ibeid, Rio Yokota, David Keyes

    International Journal of High Performance Computing Applications   30 ( 4 )   423 - 437   2016年6月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1177/1094342016634819

    Web of Science

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1405.6362v1

  • FMMの性能の可搬性 招待 査読

    横田 理央

    計算工学講演会論文集 Proceedings of the Conference on Computational Engineering and Science   21   3p   2016年5月

     詳細を見る

    記述言語:日本語   出版者・発行元:日本計算工学会  

    researchmap

  • Fast Multipole Method as a Matrix-Free Hierarchical Low-Rank Approximation 査読

    Rio Yokota, Huda Ibeid, David Keyes

    Lecture Notes in Computational Science and Engineering   117   267 - 286   2016年2月

     詳細を見る

    記述言語:英語   出版者・発行元:Springer Verlag  

    DOI: 10.1007/978-3-319-62426-6_17

    Scopus

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1602.02244v1

  • FMMとH²(HSS)行列のトレードオフについて (特集 積分方程式の近似解法 : 高速多重極法・H行列法とその応用) 査読

    横田 理央

    計算工学   21 ( 4 )   3498 - 3501   2016年

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(学術雑誌)   出版者・発行元:日本計算工学会  

    CiNii Books

    researchmap

  • Scaling FMM with data-driven OpenMP tasks on multicore architectures 査読

    Amer, A., Matsuoka, S., Pericàs, M., Maruyama, N., Taura, K., Rio Yokota, Balaji, P.

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   9903 LNCS   156 - 170   2016年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1007/978-3-319-45550-1_12

    Web of Science

    researchmap

  • Preconditioning Sparse Matrices Using a Highly Scalable Fast Multipole Method 査読

    3rd International Workshops on Advances in Computational Mechanics   2015年10月

     詳細を見る

  • Multi-Level Restricted Maximum Likelihood Covariance Estimation and Kriging for Large Non-Gridded Spatial Datasets 査読

    Julio E. Castrillon-Candas, Marc G. Genton, Rio Yokota

    Spatial Statistics   18 ( 18 )   105 - 124   2015年4月

     詳細を見る

  • Fast Multipole Preconditioners for Sparse Linear Solvers 査読

    Huda Ibeid, Rio Yokota, Jennifer Pestana, David Keyes

    Proceedings of the 11th World Congress on Computational Mechanics   2014年7月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Communication Complexity of the Fast Multipole Method and its Algebraic Variants 招待 査読

    Rio Yokota, George Turkiyyah, David Keyes

    Supercomputing Frontiers and Innovations   1 ( 1 )   63 - 84   2014年6月

     詳細を見る

    記述言語:英語  

    A combination of hierarchical tree-like data structures and data access
    patterns from fast multipole methods and hierarchical low-rank approximation of
    linear operators from H-matrix methods appears to form an algorithmic path
    forward for efficient implementation of many linear algebraic operations of
    scientific computing at the exascale. The combination provides asymptotically
    optimal computational and communication complexity and applicability to large
    classes of operators that commonly arise in scientific computing applications.
    A convergence of the mathematical theories of the fast multipole and H-matrix
    methods has been underway for over a decade. We recap this mathematical
    unification and describe implementation aspects of a hybrid of these two
    compelling hierarchical algorithms on hierarchical distributed-shared memory
    architectures, which are likely to be the first to reach the exascale. We
    present a new communication complexity estimate for fast multipole methods on
    such architectures. We also show how the data structures and access patterns of
    H-matrices for low-rank operators map onto those of fast multipole, leading to
    an algebraically generalized form of fast multipole that compromises none of
    its architecturally ideal properties.

    DOI: 10.14529/jsfi140104

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1406.1974v1

  • Petascale molecular dynamics simulation using the fast multipole method on K computer 査読

    大野 洋介, Ohno Yousuke, 横田 理央, Yokota Rio, 小山 ひろし, Koyama Hiroshi, 森本 元太郎, Morimoto Gentaro, 長谷川 亜樹, Hasegawa Aki, 舛本 現, Masumoto Gen, 沖本 憲明, Okimoto Noriaki, 平野 秀典, Hirano Yoshinori, Ibeid Huda, Ibeid Huda, 成見 哲, Narumi Tetsu, 泰地 真弘人, Taiji Makoto

    Computer Physics Communications   185 ( 10 )   2575 - 2585   2014年6月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1016/j.cpc.2014.06.004

    Web of Science

    researchmap

  • N-Body Methods

    Yokota, R., AbdulJabbar, M.

    High Performance Parallelism Pearls: Multicore and Many-core Programming Approaches   2014年

     詳細を見る

    掲載種別:研究論文(学術雑誌)   出版者・発行元:High Performance Parallelism Pearls: Multicore and Many-core Programming Approaches  

    DOI: 10.1016/B978-0-12-802118-7.00010-8

    Scopus

    researchmap

  • Fork-join and data-driven execution models on multi-core architectures: Case study of the FMM 査読

    Abdelhalim Amer, Naoya Maruyama, Miquel Pericàs, Kenjiro Taura, Rio Yokota, Satoshi Matsuoka

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   7905   255 - 266   2013年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-642-38750-0_19

    Scopus

    researchmap

  • An FMM Based on Dual Tree Traversal for Many-core Architectures 査読

    Rio Yokota

    Journal of Algorithms and Computational Technology   7 ( 3 )   301 - 324   2012年9月

     詳細を見る

    記述言語:英語  

    The present work attempts to integrate the independent efforts in the fast
    N-body community to create the fastest N-body library for many-core and
    heterogenous architectures. Focus is placed on low accuracy optimizations, in
    response to the recent interest to use FMM as a preconditioner for sparse
    linear solvers. A direct comparison with other state-of-the-art fast N-body
    codes demonstrates that orders of magnitude increase in performance can be
    achieved by careful selection of the optimal algorithm and low-level
    optimization of the code. The current N-body solver uses a fast multipole
    method with an efficient strategy for finding the list of cell-cell
    interactions by a dual tree traversal. A task-based threading model is used to
    maximize thread-level parallelism and intra-node load-balancing. In order to
    extract the full potential of the SIMD units on the latest CPUs, the inner
    kernels are optimized using AVX instructions. Our code -- exaFMM -- is an order
    of magnitude faster than the current state-of-the-art FMM codes, which are
    themselves an order of magnitude faster than the average FMM code.

    DOI: 10.1260/1748-3018.7.3.301

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1209.3516v3

  • Petascale turbulence simulation using a highly parallel fast multipole method 査読

    横田 理央, Yokota Rio, Barba Lorena, Barba Lorena, 成見 哲, Narumi Tetsu, 泰岡 顕治, Yasuoka Kenji

    Computer Physics Communications   184 ( 3 )   445 - 455   2012年9月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1016/j.cpc.2012.09.011

    Web of Science

    Scopus

    researchmap

  • 4096GPUを用いた4096³規模の一様等方性乱流の渦法解析

    横田 理央, Barba Lorena, 成見 哲

    Tsubame ESJ. : e-science journal   6   1 - 6   2012年7月

     詳細を見る

    記述言語:日本語   出版者・発行元:東京工業大学学術国際情報センター  

    researchmap

  • Data-Driven Fast Multipole Method on Distributed Memory Systems with Hardware Accelerators 査読

    Hatem Ltaief, Rio Yokota

    21st International Conference on Domain Decomposition Methods   2012年6月

     詳細を見る

    担当区分:最終著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Scalable Force Directed Graph Layout Algorithms Using Fast Multipole Methods 査読

    Enas Yunis, Rio Yokota, Aron Ahmadia

    Proceedings of the 11th International Symposium on Parallel and Distributed Computing   2012年6月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • A Parallel Numerical Simulation of Dust Particles Using Direct Numerical Simulation 査読

    Hoang Vu Nguyen, Rio Yokota, Georgiy Stenchikov

    Proceedings of the European Geosciences Union General Assembly   2012年4月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Data-Driven Execution of Fast Multipole Methods 査読

    Hatem Ltaief, Rio Yokota

    Concurrency and Computation: Practice and Experience   26 ( 11 )   1935 - 1946   2012年3月

     詳細を見る

  • Optimization of Molecular Dynamics Core Program on the K computer 査読

    Yousuke Ohno, Rio Yokota, Hiroshi Koyama, Gentaro Morimoto, Aki Hasegawa, Gen Masumoto, Tetsu Narumi, Makoto Taiji

    Proceedings of JSST 2012 International Conference on Simulation Technology   2012年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • GPU上でのFMMの木構造の構築について 査読

    横田 理央

    アンサンブル   14 ( 2 )   85 - 89   2012年

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(学術雑誌)   出版者・発行元:分子シミュレーション研究会  

    Treecode やFMM などのアルゴリズムはN 個の粒子同士の相互作用の計算量を精度と引き換えにO(N2) からO(N logN) もしくはO(N) に軽減することができる.これらの手法では,粒子をセルに分割し木構造を構築した後,その木構造を走査しながらセル単位で相互作用を計算する.このとき,木構造の走査は並列化が容易でありGPU 上での計算に向いているが,木構造の構築はGPU への実装の際に多少の工夫を要する.ここでは,GPU 上での木構造の構築の一例を示すとともに,性能向上に必要なアルゴリズム上の工夫に関する検討を行う.

    DOI: 10.11436/mssj.14.85

    researchmap

  • A Task Parallel Implementation of Fast Multipole Methods 査読

    Kenjiro Taura, Jun Nakashima, Rio Yokota, Naoya Maruyama

    2012 SC COMPANION: HIGH PERFORMANCE COMPUTING, NETWORKING, STORAGE AND ANALYSIS (SCC)   617 - 625   2012年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1109/SC.Companion.2012.86

    Web of Science

    researchmap

  • Scalable Fast Multipole Methods for Vortex Element Methods 査読

    Qi Hu, Nail A. Gumerov, Rio Yokota, Lorena Barba, Ramani Duraiswami

    2012 SC COMPANION: HIGH PERFORMANCE COMPUTING, NETWORKING, STORAGE AND ANALYSIS (SCC)   2012年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    Web of Science

    researchmap

  • Scalable Fast Multipole Methods for Vortex Element Methods 査読

    Qi Hu, Nail A. Gumerov, Rio Yokota, Lorena Barba, Ramani Duraiswami

    2012 SC COMPANION: HIGH PERFORMANCE COMPUTING, NETWORKING, STORAGE AND ANALYSIS (SCC)   1408 - +   2012年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    Web of Science

    researchmap

  • FMMを用いたペタスケール乱流解析

    横田 理央, 成見 哲, L.A.Barba, 泰岡 顕治

    情報処理学会研究報告. 計算機アーキテクチャ研究会報告   2011 ( 29 )   1 - 8   2011年11月

     詳細を見る

    記述言語:日本語   出版者・発行元:一般社団法人情報処理学会  

    Fast Multipole Method (FMM) は従来粒子のN体問題の高速化手法として発展してきたが,近年その応用の幅を広げる研究が多くなされている.本研究では,大規模 GPU システム向けに開発された FMM を用いて 20483 規模の乱流解析を行い,同様の計算条件のもとでスペクトル法との比較を行った.ただし,今回の解析に用いた手法は Treecode と FMM の長所を組み合わせたハイブリッド型になっており,GPU 上で高い Flops が出る treecode の特長をさらに高速なアルゴリズムである FMM で実現している.TSUBAME2.0 上で 4096 GPU を用いた計算において 74% の並列化効率を得た.また,このときの演算性能は 1.01PFlops であった.

    CiNii Books

    researchmap

  • FMM-based vortex method for simulation of isotropic turbulence on GPUs, compared with a spectral method 査読

    Rio Yokota, L. A. Barba

    Computers and Fluids   80 ( 80 )   17 - 27   2011年10月

     詳細を見る

  • Hierarchical N-body simulations with auto-tuning for heterogeneous systems 査読

    Rio Yokota, Lorena A. Barba

    Computing in Science and Engineering   14 ( 3 )   30 - 39   2011年8月

     詳細を見る

  • A Tuned and Scalable Fast Multipole Method as a Preeminent Algorithm for Exascale Systems 査読

    Rio Yokota, Lorena Barba

    International Journal of High Performance Computing Applications   26 ( 4 )   337 - 346   2011年6月

     詳細を見る

  • Comparing the treecode with FMM on GPUs for vortex particle simulations of a leapfrogging vortex ring 査読

    Rio Yokota, L. A. Barba

    COMPUTERS & FLUIDS   45 ( 1 )   155 - 161   2011年6月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1016/j.compfluid.2010.11.029

    Web of Science

    researchmap

  • Fast Multipole Method vs. Spectral Methods for the Simulation of Isotropic Turbulence on GPUs 査読

    Rio Yokota, Lorena Barba

    Proceedings of the 23rd International Conference on Parallel Computational Fluid Dynamics   2011年5月

     詳細を見る

    担当区分:筆頭著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Vortex methods for the simulation of turbulent flows 査読

    横田 理央, 小尾 晋之介

    Journal of Fluid Science and Technology   6 ( 1 )   14 - 29   2011年1月

     詳細を見る

  • 長崎大学大規模GPUクラスタによるN体問題とFMM 査読

    横田 理央, 濱田 剛

    計算工学   15 ( 4 )   2416 - 2419   2010年10月

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(学術雑誌)   出版者・発行元:日本計算工学会  

    CiNii Books

    researchmap

  • Biomolecular electrostatics using a fast multipole BEM on up to 512 GPUs and a billion unknowns 査読

    Rio Yokota, Jaydeep P. Bardhan, Matthew G. Knepley, L. A. Barba, Tsuyoshi Hamada

    COMPUTER PHYSICS COMMUNICATIONS   182 ( 6 )   1272 - 1283   2010年7月

     詳細を見る

  • Performance of the Fast Multipole Method on GPUs Using Various Kernels 査読

    Rio Yokota, Lorena Barba

    Proceedings of the 9th World Congress on Computational Mechanics   2010年7月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Comparing vortex methods and finite difference methods in a homogeneous turbulent shear flow 査読

    R. Yokota, S. Obi

    INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS   63 ( 7 )   828 - 846   2010年7月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(学術雑誌)  

    DOI: 10.1002/fld.2102

    Web of Science

    researchmap

  • PetRBF–A parallel O(N) algorithm for radial basis function interpolation with Gaussians 査読

    横田 理央, Yokota Rio, Barba Lorena, Barba Lorena, Knepley Matthew, Knepley Matthew

    Computer Methods in Applied Mechanics and Engineering   199 ( 25 )   1793 - 1804   2009年9月

     詳細を見る

  • Fast Multipole Methods on GPUs for the Meshfree Simulation of Turbulence 査読

    Rio Yokota, Tetsu Narumi, Ryuji Sakamaki, Kenji Yasuoka, Shinnosuke Obi

    Proceedings of the 10th US National Congress on Computational Mechanics   2009年7月

     詳細を見る

    担当区分:筆頭著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • DNS of Homogeneous Turbulence Using Vortex Methods Accelerated by the FMM on a Cluster of GPUs 査読

    Rio Yokota, Tetsu Narumi, Ryuji Sakamaki, Shun Kameoka, Kenji Yasuoka, Shinnosuke Obi

    Proceedings of the 21st International Conference on Parallel Compuational Fluid Dynamics   2009年5月

     詳細を見る

    担当区分:筆頭著者  

    researchmap

  • 平行平板間乱流における渦法の検証(解析・予測・制御 乱流の予測とモデリング(1),一般講演)

    横田 理央, 小尾 晋之介

    日本流体力学会年会講演論文集   2009   253 - 253   2009年

     詳細を見る

    記述言語:日本語   出版者・発行元:日本流体力学会  

    The vortex method is applied to the calculation of a turbulent channel flow of Re_b=5600, and the results are compared with a finite difference calculation. The fast multipole method was modified for the two way periodic boundary condition. The particle strength exchange was selected as the viscous diffusion scheme. The wall vorticity flux is calculated exactly, using a Neumann condition for the vorticity equation at the wall. The mean velocity profile agrees quantitatively between the vortex method and finite difference method.

    CiNii Books

    researchmap

  • Lagrangian Vortex Methods in Turbulent Channel Flows 査読

    R. Yokota, K. Fukagata, S. Obi

    ADVANCES IN TURBULENCE XII - PROCEEDINGS OF THE 12TH EUROMECH EUROPEAN TURBULENCE CONFERENCE   132   893 - 893   2009年

     詳細を見る

    記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    DOI: 10.1007/978-3-642-03085-7_214

    Web of Science

    researchmap

  • 2002 CIP-Level Set法を用いたマイクロ流路中の細胞挙動の数値シミュレーション(OS20.細胞・生体分子の計算バイオメカニクス(1),オーガナイズドセッション) 査読

    田村 秀一, 横田 理央, 深潟 康二

    計算力学講演会講演論文集   2009 ( 0 )   546 - 547   2009年

     詳細を見る

    記述言語:日本語   出版者・発行元:一般社団法人 日本機械学会  

    We developed a numerical method based on the Level Set method, which can simulate biological cell behavior in micro channels. The isotropic compliant wall model is used to model the cell membrane and the surface force is calculated from the local displacement of membrane computed by using the Level Set function. The biological cell model behavior is investigated in a channel with sudden expansion and contraction. The shape of the cell model is found to be elongated in the contracted region and semicircle in the expanded region. The collision between two cell models in a T-junction is also investigated. When the cell models touch with each other, their boundaries overlap. A repulsion model is proposed to avoid this overlap.

    DOI: 10.1299/jsmecmd.2009.22.546

    CiNii Books

    researchmap

  • Meshfree Simulation of Turbulence Using the Fast Multipole Methods on GPUs 査読

    Rio Yokota, Tetsu Narumi, Ryuji Sakamaki, Shun Kameoka, Kenji Yasuoka, Shinnosuke Obi

    Proceedings of the 22nd Symposium on Computational Fluid Dynamics   2008年12月

     詳細を見る

    担当区分:筆頭著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Direct Numerical Simulation of Homogeneous Shear Flow Using Vortex Methods 査読

    Rio Yokota, Shinnosuke Obi

    Proceedings of the 4th International Conference on Vortex Flows and Vortex Models   2008年4月

     詳細を見る

    担当区分:筆頭著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • Mesh-Free Simulation of the Homogeneous Shear Flow Using Vortex Methods 査読

    Rio Yokota, Shinnosuke Obi

    Proceedings of the 23rd IIS Turbulence and Shear Flow Dynamics Symposium   2008年3月

     詳細を見る

    担当区分:筆頭著者   記述言語:英語   掲載種別:研究論文(国際会議プロシーディングス)  

    researchmap

  • The study of colliding vortex rings using a special-purpose computer and FMM 査読

    SHEEL Tarun, YOKOTA Rio, YASUOKA Kenji, OBI Shinnosuke

    Transactions of the Japan Society for Computational Engineering and Science   20080003   20080003   2008年2月

     詳細を見る

    記述言語:英語  

    researchmap

  • 渦法を用いた平行平板間乱流の解析(乱流の予測とモデリング(1),一般講演)

    横田 理央, 深潟 康二, 小尾 晋之介

    日本流体力学会年会講演論文集   2008   337 - 337   2008年

     詳細を見る

    記述言語:日本語   出版者・発行元:日本流体力学会  

    The vortex method is applied to the calculation of a turbulent channel flow of Re_b=5600, and the results are compared with a finite difference calculation. The mean velocity profile agrees quantitatively between the vortex method and finite difference method for the duration of one walkout time.

    CiNii Books

    researchmap

  • Calculation of isotropic turbulence using a pure Lagrangian vortex method 査読

    横田 理央, Yokota Rio, Sheel Tarun Kumar, Sheel Tarun Kumar, 小尾 晋之介, Obi Shinnosuke

    Journal of Computational Physics   226 ( 2 )   1589 - 1606   2007年10月

     詳細を見る

  • Pure Lagrangian vortex methods for the simulation of decaying isotropic turbulence 査読

    YOKOTA RIO, 小尾 晋之介

    Proceedings of 5th Int. Symp. Turbulent Shear Flow Phenomena   2007年8月

     詳細を見る

    記述言語:英語   掲載種別:研究論文(研究会,シンポジウム資料等)  

    一様等方性乱流場を渦法により計算し、従来行なわれることのなかったDNSとの直接的な比較を通じて、双方の手法の比較を行なうとともに、渦法でこれまで評価が十分にされなかったエネルギー保存性などについて検討を加えた。

    researchmap

  • 渦法による一様せん断流の解析(乱流の予測とモデリング(4),一般講演)

    横田 理央, 小尾 晋之介

    日本流体力学会年会講演論文集   2007   195 - 195   2007年

     詳細を見る

    記述言語:日本語   出版者・発行元:日本流体力学会  

    The vortex method is applied to the calculation of a homogeneous shear flow of Re_λ=25, and the results are compared with a finite difference calculation. The fast multipole method was modified for the shear periodic boundary condition. The core spreading method and particle strength exchange were selected as the viscous diffusion scheme. The time evolution of the energy spectrum, kinetic energy and enstrophy are shown for these different cases. The component energy ratio and particle density distribution were also examined for all cases.

    CiNii Books

    researchmap

  • 662 渦法によるメッシュフリー乱流解析(OS2-1 乱流のモデリングとシミュレーション)

    横田 理央, 小尾 晋之介

    東海支部総会講演会講演論文集   2007 ( 0 )   323 - 324   2007年

     詳細を見る

    記述言語:日本語   出版者・発行元:一般社団法人 日本機械学会  

    DOI: 10.1299/jsmetokai.2007.56.323

    CiNii Books

    researchmap

  • 1202 3次元渦法・境界要素法による流体-固体連成解析(1)(OS12-1 飛ぶもの・泳ぐものとそのメカニズム・流体力学,OS12飛ぶもの・泳ぐものとそのメカニズム・流体力学)

    横田 理央, 小尾 晋之介

    流体工学部門講演会講演論文集   2006 ( 0 )   _1202 - a_   2006年

     詳細を見る

    記述言語:日本語   出版者・発行元:一般社団法人 日本機械学会  

    Fluid structure interaction of a circular cylinder is simulated by the coupling of a Vortex Element Method and Boundary Element Method. Both methods are volume-mesh-free, and thus alleviates the burden on handeling moving boundary problems in general. The rigid vibration of the cylinder is examined first, and compared with experimental studies. Then, the elastic deformation is considered.

    DOI: 10.1299/jsmefed.2006._1202-a_

    CiNii Books

    researchmap

  • 2016 渦法を用いた物体後流の3次元解析(S04-2 FEM/メッシュフリー法/粒子法とその関連技術(2),S04 FEM/メッシュフリー法/粒子法とその関連技術)

    横田 理央, 小尾 晋之介

    年次大会講演論文集   2006 ( 0 )   31 - 32   2006年

     詳細を見る

    記述言語:日本語   出版者・発行元:一般社団法人 日本機械学会  

    The wake of a bluff body is calculated using a 3-D Vortex Method and turbulence statistics are calculated from the results, which are then compared with measurements by Particle Image Velocimetry and a URANS calculation. It is shown that when compared to a 2-D Vortex Method calculation, which was previously performed by the authors, the results of the 3-D calculation is much closer to the PIV measurements and URANS calculation.

    DOI: 10.1299/jsmemecjo.2006.1.0_31

    CiNii Books

    researchmap

  • 複数の鈍い形状物体周りの渦流れシミュレーション(OS26b 乱流,反応流,混相流などの解析モデルと数値手法)

    横田 理央, 外海 憲彦, 小尾 晋之介

    計算力学講演会講演論文集   2004 ( 0 )   731 - 732   2004年

     詳細を見る

    記述言語:日本語   掲載種別:研究論文(研究会,シンポジウム資料等)   出版者・発行元:一般社団法人 日本機械学会  

    乱流の数値予測法の比較検証を目的として、一様流中に置かれた非流線型物体周りの流れを渦法と非定常RANSで計算した。2次元の計算に関しては、計算精度と時間のいずれの面でも両者の間には大きな差は見られなかった。

    DOI: 10.1299/jsmecmd.2004.17.731

    CiNii Books

    researchmap

▼全件表示

書籍等出版物

  • High Performance Parallelism Pearls

    YOKOTA Rio( 担当: 共編者(共編著者) 範囲: N-body methods on Xeon Phi coprocessors)

    Morgan Kaufmann  2014年11月 

     詳細を見る

  • GPU Computing Gems

    Morgan Kaufmann  2011年  ( ISBN:0123849888

     詳細を見る

  • GPU Computing GEMS

    Morgan Kaufmann  2011年  ( ISBN:0123849888

     詳細を見る

MISC

  • マルチプロセス実行によるGPU演算性能向上への試み

    大島聡史, 伊田明弘, 河合直聡, 深谷猛, 横田理央, 山崎市太郎

    計算工学講演会論文集(CD-ROM)   29   2024年

     詳細を見る

  • CUDA Fortran+MIG+UVMを用いたBLR行列QR分解の大規模高速化

    大島聡史, 伊田明弘, 河合直聡, 横田理央, 山崎市太郎

    情報処理学会研究報告(Web)   2023 ( HPC-190 )   2023年

     詳細を見る

  • マルチインスタンスGPU上におけるBLR行列のQR分解

    大島聡史, 伊田明弘, 横田理央, 山崎市太郎

    日本応用数理学会年会講演予稿集(CD-ROM)   2022   2022年

     詳細を見る

  • 確率的重み付け平均法のラージバッチ学習における有用性の検証

    所畑, 貴大, 長沼, 大樹, 横田, 理央

    第82回全国大会講演論文集   2020 ( 1 )   359 - 360   2020年2月

     詳細を見る

    記述言語:日本語  

    近年の深層ニューラルネットワークモデルの学習には膨大なパラメータやデータを用いるため、学習時間が増加する傾向にあり、学習の高速化が喫緊の課題である。単純に一度に用いるデータ量を増やすことで高速化を図るラージバッチ学習では、スモールバッチ学習でのNoiseの影響が少なくなるため、汎化性能の低いSharpな解への収束へと陥ることが経験的に示されている。本研究では、モデルのパラメータを確率的に平均化する手法であるSWA(Stochastic Weight Averaging)をラージバッチ学習に適用することで、汎化性能劣化問題への改善効果を検証する。

    CiNii Books

    CiNii Research

    researchmap

  • 大規模並列深層学習のための目的関数の平滑化

    長沼, 大樹, 横田, 理央

    第81回全国大会講演論文集   2019 ( 1 )   315 - 316   2019年2月

     詳細を見る

    記述言語:日本語  

    深層学習では極めて膨大な学習データを用いて学習することで他の機械学習手法を圧倒する高い性能を発揮している一方, その膨大な計算時間のため,大規模並列化によって学習時間を短縮するのが喫緊の課題である. 深層学習における問題は訓練データとの誤差を表す関数の最小化問題に帰結するが, 近年の研究によって,大規模並列化に伴うバッチサイズの増加により得られる学習モデルの汎化性能が劣化することが示されている. 本研究ではこの問題の解決方法として目的関数に対する平滑化に着目し, バッチサイズの増加を伴っても汎化性能を劣化させない目的関数の平滑化手法について検証を行う.

    CiNii Books

    CiNii Research

    researchmap

  • GPUによる階層型行列計算法の高速化に向けた多数の小密行列ベクトル積計算の最適化

    大島聡史, 山崎市太郎, 伊田明弘, 横田理央

    日本応用数理学会年会講演予稿集(CD-ROM)   2019   2019年

     詳細を見る

  • 階層型行列計算におけるソフトウェア自動チューニング

    大島 聡史, 山崎 市太郎, 伊田 明弘, 横田 理央

    計算工学講演会論文集 Proceedings of the Conference on Computational Engineering and Science   23   2018年6月

     詳細を見る

    記述言語:日本語   出版者・発行元:日本計算工学会  

    J-GLOBAL

    researchmap

  • Preface

    Rio Yokota, Michèle Weiland, John Shalf, Sadaf Alam

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   11203 LNCS   V   2018年

     詳細を見る

  • Preface

    Rio Yokota, Weigang Wu

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   10776 LNCS   V - VI   2018年

     詳細を見る

  • Preface

    Rio Yokota, Michèle Weiland, David Keyes, Carsten Trinitis

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   10876 LNCS   V - VI   2018年

     詳細を見る

  • 低ランクテンソル分解を用いた畳み込みニューラルネットワークの高速化 (パターン認識・メディア理解)

    大沢 和樹, 関谷 翠, 長沼 大樹, 横田 理央

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   117 ( 238 )   1 - 6   2017年10月

     詳細を見る

    記述言語:日本語   出版者・発行元:電子情報通信学会  

    CiNii Books

    researchmap

  • 深層学習における低精度演算を用いた高速化及びアクセラレーターの性能評価 (パターン認識・メディア理解) 査読

    長沼 大樹, 関谷 翠, 大沢 和樹, 大友 広幸, 桑村 裕二, 横田 理央

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   117 ( 238 )   101 - 107   2017年10月

     詳細を見る

    担当区分:最終著者   記述言語:日本語   出版者・発行元:電子情報通信学会  

    CiNii Books

    researchmap

  • 畳み込みニューラルネットワークの低ランク近似を用いた高速化

    大沢 和樹, 関谷 翠, 長沼 大樹, 横田 理央

    計算工学講演会論文集 Proceedings of the Conference on Computational Engineering and Science   22   5p   2017年5月

     詳細を見る

    記述言語:日本語   出版者・発行元:日本計算工学会  

    researchmap

  • 階層型行列計算のGPU向け最適化

    大島聡史, 山崎市太郎, 伊田明弘, 横田理央

    日本応用数理学会年会講演予稿集(CD-ROM)   2017   2017年

     詳細を見る

  • A Matrix-free Preconditioner for the Helmholtz Equation based on the Fast Multipole Method

    Huda Ibeid, Rio Yokota, David Keyes

    2016年8月

     詳細を見る

    Fast multipole methods (FMM) were originally developed for accelerating
    $N$-body problems for particle-based methods. FMM is more than an $N$-body
    solver, however. Recent efforts to view the FMM as an elliptic Partial
    Differential Equation (PDE) solver have opened the possibility to use it as a
    preconditioner for a broader range of applications. FMM can solve Helmholtz
    problems with optimal $\mathcal{O}(N \log N)$ complexity, has compute-bound
    inner kernels, and highly asynchronous communication patterns. The combination
    of these features makes FMM an interesting candidate as a preconditioner for
    sparse solvers on architectures of the future. The use of FMM as a
    preconditioner allows us to use lower order multipole expansions than would be
    required as a solver because individual solves need not be accurate. This
    reduces the amount of computation and communication significantly and makes the
    time-to-solution competitive with state-of-the-art preconditioners.
    Furthermore, the high asynchronicity of FMM allows it to scale to much larger
    core counts than factorization-based and multilevel methods. We describe our
    tests in reproducible details with freely available codes.

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1608.02461v1

  • Asynchronous Execution of the Fast Multipole Method Using Charm++

    Mustafa AbdulJabbar, Rio Yokota, David Keyes

    2014年5月

     詳細を見る

    Fast multipole methods (FMM) on distributed mem- ory have traditionally used
    a bulk-synchronous model of com- municating the local essential tree (LET) and
    overlapping it with computation of the local data. This could be perceived as
    an extreme case of data aggregation, where the whole LET is communicated at
    once. Charm++ allows a much finer control over the granularity of
    communication, and has a asynchronous execution model that fits well with the
    structure of our FMM code. Unlike previous work on asynchronous fast N-body
    methods such as ChaNGa and PEPC, the present work performs a direct comparison
    against the traditional bulk-synchronous approach and the asynchronous approach
    using Charm++. Furthermore, the serial performance of our FMM code is over an
    order of magnitude better than these previous codes, so it is much more
    challenging to hide the overhead of Charm++.

    arXiv

    researchmap

    その他リンク: http://arxiv.org/pdf/1405.7487v1

  • Towards a Dataflow FMM using the OmpSs Programming Model

    Miquel Pericas, Abdelhalim Amer, Keisuke Fukuda, Naoya Maruyama, Rio Yokota, Satoshi Matsuoka

    研究報告ハイパフォーマンスコンピューティング(HPC)   2012 ( 12 )   1 - 7   2012年9月

     詳細を見る

    記述言語:英語  

    This paper describes initial efforts towards the development of a dataflow implementation of the ExaFMM Fast Multipole Method code using the OmpSs programming model. We first develop several implementations based on task decomposition which overcome load balancing problems previously identified using traditional parallelization approaches. We then add dataflow extensions to improve task throughput by extracting distant parallelism and removing barriers. Execution profiles and scalability results for a single node of the Tsubame 2.0 supercomputer are then shown.This paper describes initial efforts towards the development of a dataflow implementation of the ExaFMM Fast Multipole Method code using the OmpSs programming model. We first develop several implementations based on task decomposition which overcome load balancing problems previously identified using traditional parallelization approaches. We then add dataflow extensions to improve task throughput by extracting distant parallelism and removing barriers. Execution profiles and scalability results for a single node of the Tsubame 2.0 supercomputer are then shown.

    CiNii Books

    researchmap

  • ExaFMMのタスク並列処理系MassiveThreadsによる並列化とその評価

    田浦 健次朗, 中島 潤, 横田 理央, 丸山 直也

    研究報告ハイパフォーマンスコンピューティング(HPC)   2012 ( 13 )   1 - 13   2012年7月

     詳細を見る

    記述言語:日本語  

    高速な FMM のオープンソース実装である ExaFMM を,タスク並列処理系 MassiveThreads を用いて並列化する方法,および得られた性能について述べる.FMM には木構造を用い,再帰呼び出しを用いれば見通しよく記述できる部分が多く有る.そのような処理の並列化を生産性よく行うには,任意の時点でタスクを生成し,それをスケーラブルに負荷分散できるタスク並列処理系が必須かつ有用である.とくに,タスク並列なしでは並列化が困難な以下の部分に焦点を当てる.(1) 粒子間の相互作用を作用反作用の法則に従うよう (従って全運動量を保存するよう) 計算する,(2) 木構造を生成する.Barcelona および Nehalem アーキテクチャ上で評価を行い,オリジナルの C++ コードにほとんどオーバーヘッドを加えずに並列化を行うことができた.計算時間の大部分を占める相互作用の計算に関しては,50 万以上の粒子で,Nehalem (24コア) で 90% 以上 Barcelona (32コア) で 75% 以上の並列化効率を得た.相互作用以外の計算は,現時点では並列化効率が思わしくなく,そのためすべてのフェーズを含めた合計時間については同条件で,それぞれ 50%,35% 程度となった.This paper describes parallelization of a fast, open source FMM implementation of ExaFMM, using a lightweight task parallel library MassiveThreads and reports its performance. FMM contains many operations that can be elegantly expressed with tree structures and recursions. Task parallelism is a useful tool to facilitate parallelization of such algorithms. In particular, we focus on how to elegantly parallelize operations that are otherwise hard to parallelize; (1) "mutual" interactions obeying Newton's law of action and reaction, thereby preserving the total momentum; (2) tree construction. Experiments are conducted on machines with Barcelona and Nehalem microarchitectures. The overhead due to parallelization was negligible. With 500K or more particles, computation of mutual interactions―the dominantly time consuming phase―achieved 90% utilization on Nehalem (24 cores) and 75% utilization on Barcelona (32 cores). As of writing, howerver, utilization of other phases are so low that the utilization as a whole were about 50% and 35% utilization, respectively, under the same conditions.

    CiNii Books

    researchmap

  • Turbulence Simulation Using 4096³ Vortex Particles on 4096 GPUs

    Yokota Rio, Barba Lorena, Narumi Tetsu

    Tsubame ESJ. : e-science journal   6   17 - 22   2012年7月

     詳細を見る

    記述言語:英語   出版者・発行元:東京工業大学学術国際情報センター  

    researchmap

▼全件表示

講演・口頭発表等

  • Matrices in Deep Neural Networks and How to Compute Them in Parallel 招待

    Rio Yokota

    IEEE CLUSTER 2022  2022年9月 

     詳細を見る

    開催年月日: 2022年9月

    記述言語:英語   会議種別:口頭発表(基調)  

    researchmap

  • 深層学習における2次最適化の汎化性能の検証

    石井央, 横田理央

    第84回情報処理学会全国大会  2022年3月 

     詳細を見る

    開催年月日: 2022年3月

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Vision Transformerにおけるバッチサイズの汎化性能への影響

    中村秋海, 横田理央

    第84回情報処理学会全国大会  2022年3月 

     詳細を見る

    開催年月日: 2022年3月

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • 42 TFlops Hierarchical N-body Simulation on GPUs with Applications in both Astrophysics and Turbulence 国際会議

    HAMADA Tsuyoshi, YOKOTA Rio, NITADORI Keigo, NARUMI Tetsu, YASUOKA Kenji, TAIJI Makoto, OGURI Kiyoshi

    Supercomputing 2009  2009年11月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Lagrangian Vortex Methods in Turbulent Channel Flows 国際会議

    横田 理央, 深潟 康二, 小尾 晋之介

    12th EUROMECH European Turbulence Conference  2009年9月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • 平行平板間乱流における渦法の検証

    横田 理央, 小尾 晋之介

    日本流体力学会年会  2009年9月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Fast Multipole Methods on GPUs for the Meshfree Simulation of Turbulence 国際会議

    YOKOTA Rio, NARUMI Tetsu, SAKAMAKI Ryuji, YASUOKA Kenji, OBI Shinnosuke

    10th US National Congress on Computational Mechanics  2009年7月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Comparing the treecode with FMM on GPUs for vortex particle simulations of a leapfrogging vortex ring

    22nd International Conference on Paral lel Compuational Fluid Dynamics  2010年 

     詳細を見る

  • Lagrangian simulation of turbulence using vortex methods

    2nd International Workshops on Advances in Computational Mechanics  2010年 

     詳細を見る

  • Range of Applications for the Fast Multipole Method on GPUs

    Accelerated Computing  2010年 

     詳細を見る

  • DNS of Homogeneous Turbulence Using Vortex Methods Accelerated by the FMM on a Cluster of GPUs 国際会議

    YOKOTA Rio, NARUMI Tetsu, SAKAMAKI, Ryuji, KAMEOKA Shun, YASUOKA Kenji, OBI Shinnosuke

    21st International Conference on Parallel Compuational Fluid Dynamics  2009年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • 42 TFlops Hierarchical N-body Simulation on GPUs with Applications in both Astrophysics and Turbulence

    Supercomputing 2009  2009年 

     詳細を見る

  • Lagrangian Vortex Methods in Turbulent Channel Flows

    12th EUROMECH European Turbulence Conference  2009年 

     詳細を見る

    会議種別:ポスター発表  

    researchmap

  • 渦法を用いた平行平板間乱流の解析 国際会議

    横田 理央, 小尾 晋之介

    流体力学会年会  2008年9月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Direct numerical simulation of homogeneous shear flow using vortex methods 国際会議

    YOKOTA Rio, OBI Shinnosuke

    4th International Conference on Vortex Flows and Vortex Models  2008年4月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • 渦法を用いた一様せん断流のメッシュフリー解析

    横田 理央, 小尾 晋之介

    第23回生研TSFDシンポジウム  2008年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Meshfree Simulation of Turbulence Using the Fast Multipole Methods on GPUs

    22nd Symposium on Computational Fluid Dynamics  2008年 

     詳細を見る

  • Validation of Vortex Methods in a Turblent Channel Flow

    Annual meeting of the JSFM  2009年 

     詳細を見る

  • Fast Multipole Methods on GPUs for the Meshfree Simulation of Turbulence

    10th US National Congress on Computational Mechanics  2009年 

     詳細を見る

  • DNS of Homogeneous Turbulence Using Vortex Methods Accelerated by the FMM on a Cluster of GPUs

    21st International Conference on Parallel Compuational Fluid Dynamics  2009年 

     詳細を見る

  • GPU上で高速多重極法を用いた乱流のメッシュフリー解析

    横田 理央, 成見 哲, 坂牧 隆司, 亀岡 駿, 泰岡 顕治, 小尾 晋之介

    第22回数値流体力学シンポジウム  2008年12月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Fast N-body Methods on Many-core and Heterogenous Systems 国際会議

    YOKOTA Rio

    International Workshop on Computational Science and Numerical Analysis  2012年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    開催地:東京  

    researchmap

  • Petascale Turbulence Simulation Using FMM 国際会議

    NARUMI Tetsu, YOKOTA Rio, BARBA Lorena, YASUOKA Kenji

    HOKKE-19  2011年11月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Petaflops Scale Turbulence Simulation on TSUBAME 2.0 国際会議

    YOKOTA Rio

    GPU@BU Workshop  2011年11月 

     詳細を見る

    記述言語:英語   会議種別:公開講演,セミナー,チュートリアル,講習,講義等  

    researchmap

  • Parameter Tuning of a Hybrid Treecode-FMM on GPUs 国際会議

    YOKOTA Rio, BARBA Lorena

    The First International Workshop on Characterizing Applications for Heterogeneous Exascale Systems  2011年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • A Parallel Numerical Simulation of Dust Particles Using Direct Numerical Simulation 国際会議

    NGUYEN Hoang, YOKOTA Rio, Stenchikov Gera

    European Geosciences Union General Assembly  2012年4月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • Large Scale Multi-GPU FMM for Bioelectrostatics

    SIAM Conference on Computational Science and Engineering  2011年 

     詳細を見る

  • Fast multipole method vs. spectral methods for the simulation of isotropic turbulence on GPUs 国際会議

    YOKOTA Rio, BARBA Lorena

    23rd International Conference on Parallel Computational Fluid Dynamics  2011年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Large Scale Multi-GPU FMM for Bioelectrostatics 国際会議

    YOKOTA Rio, BARBA Lorena

    Presentations SIAM Conference on Computational Science and Engineering  2011年2月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • 12 Steps to a Fast Multipole Method on GPUs 国際会議

    YOKOTA Rio

    Pan-American Advanced Studies Institute  2011年1月 

     詳細を見る

    記述言語:英語   会議種別:公開講演,セミナー,チュートリアル,講習,講義等  

    researchmap

  • Parameter Tuning of a Hybrid Treecode-FMM on GPUs

    The First International Workshop on Characterizing Applications for Heterogeneous Exascale Systems  2011年 

     詳細を見る

  • Performance of the fast multipole method on GPUs using various kernels 国際会議

    YOKOTA Rio, BARBA Lorena

    9th World Congress on Computational Mechanics  2010年7月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • RBF interpolation using Gaussians with domain decomposition on GPUs 国際会議

    YOKOTA Rio, BARBA Lorena

    SIAM annual meeting  2010年7月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Comparing the treecode with FMM on GPUs for vortex particle simulations of a leapfrogging vortex ring 国際会議

    YOKOTA Rio, BARBA Lorena

    22nd International Conference on Paral lel Compuational Fluid Dynamics  2010年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Lagrangian simulation of turbulence using vortex methods 国際会議

    YOKOTA Rio, OBI Shinnosuke

    2nd International Workshops on Advances in Computational Mechanics  2010年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • 12 Steps to a Fast Multipole Method on GPUs

    Pan-American Advanced Studies Institute  2011年 

     詳細を見る

  • (Really) Fast macromolecular electrostatics – fast algorithms, open software and accelerated computing 国際会議

    YOKOTA Rio, BARDHAN Jaydeep, KNEPLEY Matt, BARBA Lorena

    ACS Division of Physical Chemistry 240th National Meeting  2010年8月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Vortex flow simulation of multipole bluff bodies

    19th Symposium on Computational Fluid Dynamics  2005年 

     詳細を見る

  • 複数の鈍い物体周りの渦流れシミュレーション

    横田 理央, 外海 憲彦, 小尾 晋之介

    第17回計算力学講演会  2004年11月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Vortex flow simulation of multipole bluff bodies

    17th Conference on Computational Mechanics  2004年 

     詳細を見る

  • 複数の鈍い形状物体周りの渦流れシミュレーション

    横田 理央, 小尾 晋之介

    第19回数値流体力学シンポジウム  2005年12月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Vortex flow simulation of multiple bluff bodies 国際会議

    YOKOTA Rio, OBI Shinnosuke

    3rd International Conference on Vortex Flows and Vortex Models  2005年11月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Vortex flow simulation of multipole bluff bodies

    3rd International Conference on Vortex Flows and Vortex Models  2005年 

     詳細を見る

  • Vortex Method Simulation of Turbulent Channel Flow

    Annual meeting of the JSFM  2008年 

     詳細を見る

  • Direct numerical simulation of homogeneous shear flow using vortex methods

    4th International Conference on Vortex Flows and Vortex Models  2008年 

     詳細を見る

  • 渦法による一様せん断流の解析

    横田 理央, 小尾 晋之介

    日本流体力学会年会  2007年8月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • 渦法によるメッシュフリー乱流解析

    横田 理央, 小尾 晋之介

    日本機械学会東海支部第56期総会・講演会  2007年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Computation of wing-tip vortex by a three-dimensional vortex method

    21st Symposium on Computational Fluid Dynamics  2007年 

     詳細を見る

  • Meshfree direct numerical simulation of turbulence using the vortex method on parallel MDGRAPE-3 boards along with the fast multipole method

    Next-Generation Supercomputing Symposium  2007年 

     詳細を見る

    会議種別:ポスター発表  

    researchmap

  • Mesh-free simulation of the homogeneous shear flow using vortex methods

    23rd IIS Turbulence and Shear Flow Dynamics Symposium  2008年 

     詳細を見る

  • 三次元渦法による翼端渦の数値解析

    佐藤 彰, 横田 理央, 小尾 晋之介

    第21回数値流体力学シンポジウム  2007年12月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • MDGRAPE-3を用いた渦法による乱流の直接数値シミュレーション

    横田 理央, 成見 哲, 泰岡 顕治, 戎崎 俊一, 小尾 晋之介

    次世代スーパーコンピューティング・シンポジウム  2007年10月 

     詳細を見る

    記述言語:日本語   会議種別:ポスター発表  

    researchmap

  • Pure Lagrangian vortex methods for the simulation of decaying isotropic turbulence 国際会議

    YOKOTA Rio, OBI Shinnosuke

    International Symposium on Turbulence and Shear Flow Phenomena  2007年8月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Calculation of the decay of colliding turbulent vortex rings

    4th International Conference on Vortex Flows and Vortex Models  2008年 

     詳細を見る

  • Pure Lagrangian vortex methods for the simulation of decaying isotropic turbulence

    5th International Symposium on Turbulence and Shear Flow Phenomena  2007年 

     詳細を見る

  • Simulation of homogeneous isotropic turbulence using the vortex method

    20th Symposium on Computational Fluid Dynamics  2006年 

     詳細を見る

  • Calculation of fluid structure interaction using VEM and BEM

    Conference of the JSME Fluid Engineering Division  2006年 

     詳細を見る

  • Simulation of a wake using 3-D vortex element method

    Annual Meeting of the JSME  2006年 

     詳細を見る

  • Vortex flow simulation between multipole bridge decks

    Whither Turbulence Prediction and Control  2006年 

     詳細を見る

  • 渦法による一様等方性乱流の解析

    横田 理央, 小尾 晋之介

    第20回数値流体力学シンポジウム  2006年12月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • 3次元渦法・境界要素法による流体-固体連成解析

    横田 理央, 小尾 晋之介

    日本機械学会流体工学部門講演会  2006年10月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • 渦法を用いた物体後流の3次元解析

    横田 理央, 小尾 晋之介

    日本機械学会年次大会  2006年9月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • Vortex flow simulation between multiple bridge decks 国際会議

    YOKOTA Rio, OBI Shinnosuke

    Whither Turbulence Prediction and Control  2006年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Vortex methods for the calculation of homogeneous shear flows

    Annual meeting of the JSFM  2007年 

     詳細を見る

  • Meshfree turbulence simulation using vortex methods

    56th Conference of the JSME Tokai Branch  2007年 

     詳細を見る

  • Performance of the fast multipole method on GPUs using various kernels

    9th World Congress on Computational Mechanics  2010年 

     詳細を見る

  • Fast N-body Methods as a Compute-Bound Preconditioner for Sparse Solvers on GPUs 国際会議

    YOKOTA Rio

    GPU Technology Conference  2014年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Fast Multipole Method Preconditioning 国際会議

    PESTANA Jennifer, YOKOTA Rio, IBEID Huda, KEYES David

    International Conference On Preconditioning Techniques For Scientific And Industrial Applications  2013年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Fork-Join and Data-Driven Execution Models on Multi-core Architectures: Case Study of the FMM 国際会議

    Abdelhalim Amer, Naoya Maruyama, Miquel Pericas, Kenjiro Taura, Rio Yokota, Satoshi Matsuoka

    International Supercomputing Conference  2013年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:ライプチヒ  

    researchmap

  • Advances in Fast Multipole Methods for Scalable Electrostatics Calculations 招待 国際会議

    YOKOTA Rio

    Workshop: Electrostatics methods in Molecular Simulation  2013年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    researchmap

  • ExaFMM – a Testbed for Comparing Various Implementations of the FMM 国際会議

    YOKOTA Rio

    SIAM Conference on Computational Science and Engineering  2015年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Fast Multipole Preconditioners for Sparse Linear Solvers 国際会議

    IBEID Huda, YOKOTA Rio, KEYES David

    11th World Congress on Computational Mechanics  2014年7月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Communication Complexity of the Fast Multipole Method and its Alge- braic Variants 国際会議

    YOKOTA Rio, KEYES David

    CBMS-NSF Conference: Fast Direct Solvers for Elliptic PDEs  2014年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • High Performance Numerical Algorithms for Seismic and Reservoir Sim- ulations 国際会議

    LTAIEF Hatem, YOKOTA Rio

    GPU Technology Conference  2014年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Various Implementations of FMM and Their Performance on Future Architectures 国際会議

    YOKOTA Rio

    Multi-resolution Interactions Workshop  2015年8月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Fast Multipole Method as Preconditioner 国際会議

    Ibeid Huda, Pestana Jennifer, 横田理央, Keyes David

    SIAM Conference on Computational Science and Engineering  2015年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:ソルトレイクシティー  

    researchmap

  • Recent Trends in Hierarchical N-body Methods on GPUs 国際会議

    YOKOTA Rio, BARBA Lorena

    GPU Technology Conference  2012年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Scaling Fast Multipole Methods up to 4000 GPUs 招待 国際会議

    YOKOTA Rio, NARUMI Tetsu, BARBA Lorena, YASUOKA Kenji

    ATIP/A*CRC Workshop on Accelerator Technologies for High Performance Computing  2012年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Running Fast Multipole Method on the Full Node of TSUBAME and K computer 国際会議

    YOKOTA Rio

    Scalable Hierarchical Algorithms for Extreme Computing  2012年4月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Petascale Fast Multipole Methods on GPUs 招待 国際会議

    YOKOTA Rio

    GPU Technology Conference Japan  2012年7月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Petascale Fast Multipole Methods on GPUs 招待 国際会議

    YOKOTA Rio

    The 11th International Symposium on Parallel and Distributed Computing  2012年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(基調)  

    researchmap

  • Scalable Force Directed Graph Layout Algorithms Using Fast Multipole Methods 国際会議

    YUNIS Enas, YOKOTA Rio, AHMADIA Aron

    The 11th International Symposium on Parallel and Distributed Computing  2012年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Data-Driven Fast Multipole Method on Distributed Memory Systems with Hardware Accelerators 国際会議

    LTAIEF Hatem, YOKOTA Rio

    21st International Conference on Domain Decomposition Methods  2012年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Investigating New Numerical Techniques for Reservoir Simulations on GPUs 国際会議

    ABDELFETTAH Ahmad, LTAIEF Hatem, YOKOTA Rio

    GPU Technology Conference  2013年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Fast Multipole Method as a Preconditioner 国際会議

    IBEID Huda, YOKOTA Rio, KEYES David

    SIAM Conference on Computational Science and Engineering  2013年2月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • A Task Parallelism Meets Fast Multipole Methods 国際会議

    TAURA Kenjiro, NAKASHIMA Jun, YOKOTA Rio, MARUYAMA, Naoya

    Workshop on Latest Advances in Scalable Algorithms for Large-Scale Systems  2012年11月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Compute-Memory Tradeoff in Hierarchical Low-Rank Approximation Methods 国際会議

    横田理央

    SIAM Conference on Computational Science and Engineering  2017年2月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:アトランタ  

    researchmap

  • Energy Conservation of Fast Multipole Methods in Classical Molecular Dynamics Simulations 招待 国際会議

    横田理央

    7th AICS International Symposium  2017年2月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    開催地:神戸  

    researchmap

  • Communication Reducing Algorithms for Distributed Heirarchical N-Body Methods 国際会議

    AbdulJabbar Mustafa, Markomanolis George, Ibeid Huda, 横田理央, Keyes David

    32nd International Conference, ISC High Performance  2017年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:フランクフルト  

    researchmap

  • 畳み込みニューラルネットワークの低ランク近似を用いた高速化

    大沢和樹, 関谷翠, 長沼大樹, 横田理央

    第22回計算工学講演会  2017年5月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:大宮  

    researchmap

  • 低ランク近似行列によるCNNにおける畳み込み演算の最適化

    本山 義史, 遠藤 敏夫, 松岡 聡, 横田 理央, 福田 圭祐, 佐藤 育郎

    第158回ハイパフォーマンスコンピューティング研究発表会  2017年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:熱海  

    researchmap

  • 低ランク近似を用いた深層学習の行列積の高速化

    関谷翠, 大沢和樹, 長沼大樹, 横田理央

    第158回ハイパフォーマンスコンピューティング研究発表会  2017年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:熱海  

    researchmap

  • Performance Evaluation of Computation and Communication Kernels of the Fast Multipole Method on Intel Manycore Architecture 国際会議

    AbdulJabbar Mustafa, Al Farhan Mohammed, 横田理央, Keyes David

    3rd International European Conference on Parallel and Distributed Computing  2017年8月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:ゲッチンゲン  

    researchmap

  • GPUクラスタ上における階層型行列計算の最適化

    大島 聡史, 山崎 市太郎, 伊田 明弘, 横田理央

    Summer United Workshops on Parallel, Distributed and Cooperative Processing  2017年7月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:秋田  

    researchmap

  • Accelerating Matrix Multiplication in Deep Learning by Using Low-Rank Approximation 国際会議

    大沢和樹, 関谷翠, 長沼大樹, 横田理央

    The 2017 International Conference on High Performance Computing & Simulation  2017年7月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:ジェノア  

    researchmap

  • Hierarchical Low-Rank Approximations at Extreme Scale 招待 国際会議

    横田理央

    32nd International Conference, ISC High Performance  2017年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    開催地:フランクフルト  

    researchmap

  • Fast Multipole Method as a Matrix-free Hierarchical Low-rank Approximation 国際会議

    YOKOTA Rio

    International Workshop on Eigenvalue Problems  2015年9月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • A Common API for Fast Multipole Methods 国際会議

    YOKOTA Rio

    Accelerate Data Analytics and Computing Workshop  2016年1月 

     詳細を見る

    記述言語:英語   会議種別:シンポジウム・ワークショップ パネル(指名)  

    researchmap

  • Tuning Parameters in FMM 招待

    YOKOTA Rio

    Seventh Symposium on Automatic Tuning Technology and its Application  2015年12月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(招待・特別)  

    researchmap

  • Comparison of FMM and HSS at Large Scale 国際会議

    YOKOTA Rio, ROUET, Francois-Henri, LI Xiaoye

    SIAM Conference on Applied Linear Algebra  2015年10月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Preconditioning Sparse Matrices Using a Highly Scalable Fast Multipole Method 国際会議

    YOKOTA Rio, IBEID Huda, KEYES David

    3rd International Workshops on Advances in Computational Mechanics  2015年10月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Improving Data Locality of Fast Multipole Methods 国際会議

    YOKOTA Rio

    Third Workshop on Programming Abstractions for Data Locality  2016年10月 

     詳細を見る

    記述言語:英語   会議種別:シンポジウム・ワークショップ パネル(公募)  

    researchmap

  • Fast Multipole Method を用いた多種アーキテクチャ向け スーパーコンピュータ用ライブラリの開発と 分子・流体シミュレーションでの評価

    横田 理央

    学際大規模情報基盤共同利用・共同研究拠点 第8回シンポジウム  2016年7月 

     詳細を見る

    記述言語:日本語   会議種別:シンポジウム・ワークショップ パネル(公募)  

    researchmap

  • Perforamance Portability of FMM

    横田 理央

    第21回計算工学講演会  2016年5月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    researchmap

  • A Matrix-Free Preconditioner for Elliptic Solvers Based on the Fast Multipole Method 国際会議

    Ibeid Huda, 横田理央, Keyes David

    SIAM Conference on Parallel Processing for Scientific Computing  2016年4月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:パリ  

    researchmap

  • Tapas: An Implicitly Parallel ProgrammingFramework For Hierarchical N-body Algorithms 国際会議

    福田 圭祐, 松田 元彦, 丸山直也, 横田 理央, 田浦 健次朗, 松岡 聡

    The 22nd IEEE International Conference on Parallel And Distributed Systems  2016年12月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:武漢  

    researchmap

  • O(N)で並列性の高い密行列のLU分解 招待

    横田理央

    RIMS 研究集会 「数値解析が拓く次世代情報社会~エッジから富岳まで~」  2022年12月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(招待・特別)  

    researchmap

  • 人工画像を用いたVision Transformerの大規模事前学習 招待

    横田理央

    社会的課題解決型データサイエンス・AI研究推進体シンポジウム  2022年9月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(招待・特別)  

    researchmap

  • 人工画像を用いたVision Transformerの大規模事前学習 招待

    横田理央

    DENSO IT LAB x TOKYO TECH Discussion Night in MIRU  2022年7月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(招待・特別)  

    researchmap

  • 大規模並列深層学習のための目的関数の平滑化

    長沼大樹, 横田理央

    第81回情報処理学会全国大会  2019年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:福岡  

    researchmap

  • Scaling Deep Learning to Thousands of GPUs 招待 国際会議

    Rio Yokota

    HPC 2018  2018年7月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    開催地:Cetraro  

    researchmap

  • Energy Conserving Fast Multipole Methods for the Calculation of Long-range Interactions 招待 国際会議

    Rio Yokota

    Mathematics in Action: Modeling and analysis in molecular biology and electro- physiology  2018年6月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    開催地:Suzhou  

    researchmap

  • Analyzing Performance of BiCGStab with Hierarchical Matrix on GPU clusters 国際会議

    Ichitaro Yamazaki, Ahmad Abdelfattah, Akihiro Ida, Satoshi Ohshima, Stanimire Tomov, Rio Yokota, Jack Dongarra

    32nd IEEE International Parallel & Distributed Processing Symposium  2018年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:バンクーバー  

    researchmap

  • Optimization of Hierarchical Matrix Computation on GPU 国際会議

    Satoshi Ohshima, Ichitaro Yamazaki, Akihiro Ida, Rio Yokota

    SC Asia  2018年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:シンガポール  

    researchmap

  • 自然勾配法に基づく変分深層学習

    中田光, 大沢和樹, 横田理央

    第81回情報処理学会全国大会  2019年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:福岡  

    researchmap

  • Optimization Methods for Large Scale Distributed Deep Learning 招待 国際会議

    Rio Yokota

    IPAM Workshop I: Big Data Meets Large-Scale Computing  2018年9月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    開催地:Los Angeles  

    researchmap

  • 自然勾配近似法を用いた大規模並列深層学習におけるハイパーパラメータ最適化

    長沼大樹, 岩瀬 駿, 郭 林昇, 中田 光, 横田 理央

    第17回情報科学技術フォーラム 2018  2018年9月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:福岡  

    researchmap

  • Early Application Results on TSUBAME 3 招待 国際会議

    Rio Yokota

    Smoky Mountains Computational Sciences and Engineering Conference  2018年8月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(招待・特別)  

    開催地:Gatlinburg  

    researchmap

  • Fisher情報行列の解析に基づく大規模深層学習のための二次最適化手法

    大沢和樹, 横田理央, Chuan-Sheng Foo, Vijay Chandrasekhar

    第81回情報処理学会全国大会  2019年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:福岡  

    researchmap

  • Tensorコアを用いたBatched QR分解

    大友広幸, 横田理央

    第81回情報処理学会全国大会  2019年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:福岡  

    researchmap

  • 低ランクテンソル分解を用いた畳み込みニューラルネットワークの高速化

    大沢和樹, 関谷翠, 長沼大樹, 横田理央

    パターン認識・メディア理解研究会  2017年10月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:熊本  

    researchmap

  • Evaluating the Compression Efficiency of the Filters in Convolutional Neural Networks 国際会議

    大沢和樹, 横田理央

    The 26th International Conference on Artificial Neural Networks  2017年9月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:サルデニア  

    researchmap

  • 深層学習における半精度演算を用いた圧縮モデルの高速化

    長沼大樹, 大沢和樹, 関谷翠, 横田理央

    日本応用数理学会年会  2017年9月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:東京  

    researchmap

  • フィッシャー情報行列のクロネッカー因子分解を用いた深層ニューラルネットワークの分散学習

    大友 広幸, 大沢 和樹, 横田 理央

    第163回ハイパフォーマンスコンピューティング研究発表会  2018年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:愛媛  

    researchmap

  • Accelerating Convolutional Neural Networks Using Low Precision Arithmetic 国際会議

    Hiroki Naganuma, Rio Yokota

    HPC Asia  2018年1月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    開催地:東京  

    researchmap

  • 畳み込みニューラル ネットワークにおける低精度演算を用いた高速化の検証

    長沼大樹, 横田理央

    GTC Japan  2017年12月 

     詳細を見る

    記述言語:日本語   会議種別:ポスター発表  

    開催地:東京  

    researchmap

  • 深層学習における低精度演算を用いた高速化及びアクセラレーターの性能評価

    長沼大樹, 関谷翠, 大沢和樹, 大友広幸, 桑村裕二, 横田理央

    パターン認識・メディア理解研究会  2017年10月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:熊本  

    researchmap

  • フィッシャー情報行列のクロネッカー因子分解を用いた深層学習

    大友広幸, 大沢和樹, 横田理央

    情報処理学会 全国大会  2018年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:東京  

    researchmap

  • Can we use Hierarchical Low-Rank Approximation for Deep Learning? 招待 国際会議

    Rio Yokota

    HPC Saudi 2018  2018年3月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(基調)  

    開催地:Jeddah  

    researchmap

  • 自然勾配法の近似手法における学習パラメータの調整

    桑村祐二, 大沢和樹, 横田理央

    情報処理学会全国大会  2018年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:東京  

    researchmap

  • Self-supervised Continual Pretraining for Class Incremental Image Classification

    Hikaru Nakata, Nakamasa Inoue, Rio Yokota

    Proc. CVPR CLVISION Workshop (Findings)  2021年6月 

     詳細を見る

    記述言語:英語  

    researchmap

  • Effect of Mixed Precision Computing on H-Matrix Vector Multiplication in BEM Analysis

    Rise Ooi, Takeshi Iwashita, Takeshi Fukaya, Akihiro Ida, Rio Yokota

    HPC Asia 2020  2020年1月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Practical Deep Learning with Bayesian Principles 国際会議

    Osawa, K, Swaroop, S, Jain, A, Eschenhagen, R, Turner, R. E, Yokota, R, Khan, M. E

    The 33rd Conference on Neural Information Processing Systems  2019年12月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    開催地:バンクーバー  

    researchmap

  • On Empirical Analysis of Layer-wised Learning Rate Schedule

    Hiroki Naganuma, Rio Yokota

    ACML 2019 Workshop on Statistics & Machine Learning Researchers  2019年11月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • TSQR on TensorCores

    Hiroyuki Ootomo, Rio Yokota

    The International Conference for High Performance Computing, Networking, Storage, and Analysis  2019年11月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • Randomized SVD on TensorCores

    Hiroyuki Ootomo, Rio Yokota

    ISC High Performance 2020  2020年6月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • Distributed Memory Task-Based Block Low Rank Direct Solver

    Sameer Deshmukh, Rio Yokota

    ISC High Performance 2020  2020年6月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • QR Decomposition of Block Low-Rank Matrices

    Muhammad Ridwan Apriansyah, Rio Yokota

    HPC Asia 2020  2020年1月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • Distributed Memory Task-Based Block Low Rank Direct Solver

    Sameer Deshmukh, Rio Yokota

    HPC Asia 2020  2020年1月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • Runtime System for GPU-based Hierarchical LU factorization

    Qianxing Ma, Rio Yokota

    The International Conference for High Performance Computing, Networking, Storage, and Analysis  2019年11月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    researchmap

  • Optimization of Numerous Small Dense-Matrix?Vector Multiplications in H-matrix Arithmetic on GPU 国際会議

    大島 聡史, 山崎 市太郎, 伊田 明弘, 横田 理央

    Auto-Tuning for Multicore and GPU (ATMG) In conjunction with the IEEE MCSoC-19  2019年10月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:シンガポール  

    researchmap

  • A Performance Improvement Approach for Second-Order Optimization in Large Mini-batch Training 国際会議

    長沼 大樹, 横田 理央

    2nd High Performance Machine Learning Workshop CCGrid2019 (HPML2019)  2019年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:ラカルナ  

    researchmap

  • Exhaustive Study of Hierarchical AllReduce Patterns for Large Messages Between GPUs 国際会議

    上野 裕一郎, 横田 理央

    19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID)  2019年5月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    開催地:ラカルナ  

    researchmap

  • ノイズ注入による平均化を用いたラージバッチ学習の汎化性能改善手法の検討

    長沼 大樹, 横田 理央

    電子情報通信学会総合大会  2019年3月 

     詳細を見る

    記述言語:日本語   会議種別:ポスター発表  

    開催地:東京  

    researchmap

  • 大規模並列深層学習における2次の最適化手法の効果 招待

    横田理央, 大沢和樹, 辻陽平, 上野裕一郎, 成瀬彰

    電子情報通信学会総合大会  2019年3月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(招待・特別)  

    開催地:東京  

    researchmap

  • Tensorコアを用いたTSQRのGPU実装

    大友 広幸, 横田 理央

    第170回ハイパフォーマンスコンピューティング研究発表  2019年7月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:北見  

    researchmap

  • Flexible and Simplistic Hierarchical Matrix-Based Fast Direct Solver

    P. Spalthoff, R. Yokota

    第170回ハイパフォーマンスコンピューティング研究発表  2019年7月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:北見  

    researchmap

  • Large-Scale Distributed Second-Order Optimization Using Kronecker-Factored Approximate Curvature for Deep Convolutional Neural Networks 国際会議

    2019年6月 

     詳細を見る

    記述言語:英語   会議種別:ポスター発表  

    開催地:ロングビーチ  

    researchmap

  • ラージバッチ学習のための自然勾配学習法におけるSmoothingの有効性

    長沼 大樹, 横田 理央

    The 3rd Cross-disciplinary Workshop on Computing Systems, Infrastructures, and Programming (xSIG)  2019年5月 

     詳細を見る

    記述言語:日本語   会議種別:口頭発表(一般)  

    開催地:横浜  

    researchmap

  • Performance Optimizations and Analysis of Distributed Deep Learning with Approximated Second-Order Optimization Method

    Yohei Tsuji, Kazuki Osawa, Yuichiro Ueno, Akira Naruse, Rio Yokota, Satoshi Matsuoka

    International Conference on Parallel Processing: The 1st Workshop on Parallel and Distributed Machine Learning, 48th International Conference on Parallel Processing  2019年8月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • Range of Applications for the Fast Multipole Method on GPUs 国際会議

    YOKOTA Rio

    Accelerated Computing  2010年1月 

     詳細を見る

    記述言語:英語   会議種別:口頭発表(一般)  

    researchmap

  • (Really) Fast macromolecular electrostatics – fast algorithms, open software and accelerated computing

    ACS Division of Physical Chemistry 240th National Meeting  2010年 

     詳細を見る

  • RBF interpolation using Gaussians with domain decomposition on GPUs

    SIAM annual meeting  2010年 

     詳細を見る

▼全件表示

受賞

  • ACM Gordon Bell Prize (price/performance)

    2009年  

     詳細を見る

共同研究・競争的資金等の研究課題

  • 次世代計算機の潜在能力を引き出すための科学技術ソフトウェアの刷新

    研究課題/領域番号:25H01109  2025年4月 - 2028年3月

    日本学術振興会  科学研究費助成事業  基盤研究(A)

    横田 理央

      詳細を見る

    配分額:46020000円 ( 直接経費:35400000円 、 間接経費:10620000円 )

    researchmap

  • 大規模言語モデル(LLM:Large Language Model)を活用した医薬品等の有効性・安全 性評価のためのアウトカム抽出の方法論の確立に向けた研究

    研究課題/領域番号:24AC0401  2024年4月 - 2027年3月

    厚生労働省  厚生労働科学研究費  応用研究

    武藤 学, 松本 繁巳, 中島 貴子, 黒田 知宏, 吉原 博幸, 小林 慎治, 粂 直人, 横田 理央, 加藤 康之

      詳細を見る

  • 低ランク構造行列法の適用範囲拡大と多様な計算アーキテクチャの活用

    研究課題/領域番号:24K02949  2024年4月 - 2027年3月

    日本学術振興会  科学研究費助成事業  基盤研究(B)

    伊田 明弘, 横田 理央, 塙 敏博, 岩下 武史, 大島 聡史, 星野 哲也, 平石 拓, 河合 直聡

      詳細を見る

    配分額:18590000円 ( 直接経費:14300000円 、 間接経費:4290000円 )

    researchmap

  • 計算科学・計算工学の未来を拓く次世代高性能線形ソルバ

    研究課題/領域番号:23H00462  2023年4月 - 2027年3月

    日本学術振興会  科学研究費助成事業  基盤研究(A)

    岩下 武史, 塙 敏博, 伊田 明弘, 美舩 健, 横田 理央, 高橋 康人, 今倉 暁, 深谷 猛

      詳細を見る

    配分額:45500000円 ( 直接経費:35000000円 、 間接経費:10500000円 )

    researchmap

  • 深層生成モデルを活用した構成的なパターン認識・理解

    研究課題/領域番号:23H00490  2023年4月 - 2026年3月

    日本学術振興会  科学研究費助成事業  基盤研究(A)

    篠田 浩一, 井上 中順, 横田 理央, 川上 玲, 佐藤 育郎

      詳細を見る

    配分額:47190000円 ( 直接経費:36300000円 、 間接経費:10890000円 )

    本研究課題では,識別の対象(インスタンス)を属性の集合(束)とみなし,特徴量空間においてその特徴を属性ごとに分解する.そして,これらの属性特徴からインスタンスを再合成する過程で属性特徴を最適化することで,各属性を高精度で識別し,かつ,外れ値に対し頑健な識別手法を実現することを目的としている。このために深層生成モデルと高密度な属性アノテーションに基づく学習手法を開発する.従来研究の多くが対象とその属性が一対一に対応する平坦な意味構造を仮定していたのに対し,本研究は多くの属性が複雑に絡み合う対象における複数の属性を同時に識別することを可能にする.新しい属性やクラスの創発も視野に入れる.より具体的には、深層学習を用いた「合成による識別」のアプローチにより,構成的なパターン認識・理解を行う方法論を確立する.人の動作認識,話者・感情認識,マルチモーダル認識の3つのタスクで横断的に評価し,従来に比べ高い識別性能を目指す.初年度である本年度は、人の動作認識、話者・感情認識、マルチモーダル認識の各々の課題において、評価データベースの構築と、ベースライン方式の開発を行った。これらと並行して、比較的小規模なタスクで、拡散モデルなどの生成モデルを用いて識別を行う方式の開発を行った。また、ニューラル構造探索などを用いて生成モデルの効率的な学習を行う方式も開発した。特に、センサーと映像のマルチモーダル認識における基本方式の構築、およびデータベース構築、人間の歩容認識の基本方式の開発、マルチモーダル感情認識の基本方式の開発を行った。

    researchmap

  • 2相乱流の普遍統計法則の解明による流体機器の混合促進機能向上

    研究課題/領域番号:22H01403  2022年4月 - 2026年3月

    日本学術振興会  科学研究費助成事業  基盤研究(B)

    小林 宏充, 横田 理央, 湯井 悟志

      詳細を見る

    配分額:17290000円 ( 直接経費:13300000円 、 間接経費:3990000円 )

    researchmap

  • 2相乱流の普遍統計法則の解明による流体機器の混合促進機能向上

    研究課題/領域番号:23K22674  2022年4月 - 2026年3月

    日本学術振興会  科学研究費助成事業  基盤研究(B)

    小林 宏充, 横田 理央, 湯井 悟志

      詳細を見る

    配分額:17290000円 ( 直接経費:13300000円 、 間接経費:3990000円 )

    本研究では、量子乱流・常流体乱流の2相乱流と通常の2相乱流は何がどう異なるのか、その際の普遍統計法則はどのようなものかを解明する。そこで、今年度の目的を、これまでの実験から得られた常流体と超流体の間の相互摩擦力のモデル係数を利用する場合と最近提案された実験結果を必要としない理論的なモデルの検証を実施することとした。本目的を達成するために、量子渦輪が収縮しながら進行する様を可視化した実験の再現を行った。実験では、量子渦輪の可視化のために重水素の固体粒子を渦輪に10個程度を付着させている。その付着粒子による相互摩擦力も加味した数値計算を実施した。
    その結果、収縮する渦輪の半径の時間変化は、理論的なモデルのほうが、実験結果をよく説明することがわかった。本結果は、実験結果に寄らない自己完結型のモデルの進展という観点で、大変重要な結果である。本結果は、Nature Communicationsに掲載され、注目を集めている。
    また、周期的な外力を与えた常流体の一様等方性乱流中で、量子乱流を発現させた場合、量子渦がバンドル状になり、-5/3乗を示すエネルギースペクトルが低波数で得られることがわかった。一方、熱対向流のように常流体と超流体の平均相対速度が存在する場合は、相互摩擦力による常流体への巻き付きよりも相対速度による移流効果が上回り、バンドルを形成しないことが分かった。このとき、先の-5/3乗則を示す準古典乱流に対して、-1乗則のエネルギースペクトルとなる。これはVinenによって提唱される超量子乱流状態であると考えられる。これらの違いが古典乱流と量子乱流の統計則の違いの一つであると考えられ、重要な知見が得られた。

    researchmap

  • 階層的低ランク近似による高速・高精度な固有値計算と大規模電子状態計算への応用

    研究課題/領域番号:22H03598  2022年4月 - 2025年3月

    日本学術振興会  科学研究費助成事業 基盤研究(B)  基盤研究(B)

    横田理央, 伊田明宏, 荻田武史, 星健夫

      詳細を見る

    担当区分:研究代表者 

    配分額:17680000円 ( 直接経費:13600000円 、 間接経費:4080000円 )

    researchmap

  • 階層的低ランク近似による高速・高精度な固有値計算と大規模電子状態計算への応用

    研究課題/領域番号:23K24854  2022年4月 - 2025年3月

    日本学術振興会  科学研究費助成事業  基盤研究(B)

    横田 理央, 荻田 武史, 星 健夫, 伊田 明弘

      詳細を見る

    配分額:17680000円 ( 直接経費:13600000円 、 間接経費:4080000円 )

    電子状態計算では密行列の固有値問題を解く必要があるが、素朴な手法ではN次元の行列に対してO(N^3)の計算量を要する。既存研究では、小さい行列要素を無視することで疎行列とみなすが、疎行列近似により本来あるべき行列の正定値性が失われ、解法が数理的に破綻することさえある。そのため,疎行列に近似することなく高速に計算する手法が確立されれば、大きなブレークスルーとなる。本研究では、密行列を階層的にブロック分割し、非対角ブロックを低ランク近似することで行列積や行列分解の O(N^3) の計算量を O(N log^2 N) あるいは O(N) にまで低減できる手法を開発する。
    2023年度には、代表者の横田らのH行列によるコレスキー分解をLDL^T分解に拡張し、固有値の二分探索を行うことでk番目の固有値をO(NlogN)で求める方法を開発した。これと並行して2023年度には、固有値分解における三重対角化に対してH行列を適用した。これにより、密行列としては扱えない巨大な行列の全固有値を計算可能であることを明らかにした。さらに、2023年度には、H行列を用いた場合でも密行列と同じ精度で固有値計算ができていることを精度保証の枠組みを利用することで証明した。ただし、このときの実験には比較的小規模な行列を用いることで、密行列の直接解法や精度保証の計算が現実的な時間で完了するように配慮した。

    researchmap

  • A New Bayes Duality Principle for Adaptive, Robust, Life-long Learning of AI

    研究課題/領域番号:JY210177nn  2021年10月 - 2027年3月

    JST  CREST 

    Emtiyaz Khan, Kenichi Bannai, Rio Yokota, Julyan Arbel

      詳細を見る

    担当区分:研究分担者 

    researchmap

  • 格子H行列に基づく数値線形代数の構築と最新アーキテクチャへの高性能実装法

    研究課題/領域番号:21H03447  2021年4月 - 2024年3月

    日本学術振興会  科学研究費助成事業  基盤研究(B)

    伊田 明弘, 横田 理央, 塙 敏博, 岩下 武史, 大島 聡史, 星野 哲也, 平石 拓

      詳細を見る

    配分額:17290000円 ( 直接経費:13300000円 、 間接経費:3990000円 )

    格子H行列を用いた数値線形代数系の構築を目的とし、固有値計算、LU分解、QR分解などを実行する新しいアルゴリズムの研究・開発を行った。多くの成果が得られた中でも、主要なものとして、BLR(Block Low-Rank)行列の固有値計算法が挙げられる。格子H行列の特殊な場合であるBLR行列について、全固有値を計算するアルゴリズムを開発した。開発したアルゴリズムの計算複雑度を、BLR行列を特徴付ける行列サイズ・ブロックサイズ・各ブロックの階数(ランク)を変数として見積り、最適な条件を検討した。最適条件下で提案アルゴリズムの計算量が従来の密行列の場合に比べて大幅に低減させられることを理論的に示した。また、数値実験を用いて、計算時間が理論通り計算量に比例すること、および、固有値と固有ベクトルの誤差が階数を増やすにつれて密行列に近づいていくことを確かめた。
    格子H行列法の適用範囲の拡大に取り組んだ。従来、格子H行列を含む低ランク構造行列法は境界要素解析(時間項を含まない空間領域の積分方程式法)への適用が想定されていた。時空間領域積分方程式法に適用範囲を拡張すべく、FDP(Fast Domain Partitioning)法と格子H行列法を組み合わせた手法を開発し、従来手法の計算量を開発手法では大幅に低減させられることを理論的に示した。さらに、3次元弾性波動伝播解析を提案手法で行う計算コードを開発し、理論に近い計算時間で解析が行えることを確かめた。
    格子H行列法の高性能実装に関する研究においても多くの成果が得られた。主要な成果としては、タスク並列言語Tascellを拡張し、分散メモリ環境における行列分割構造生成の並列実装を提案した。3次元電場解析に対する約1億要素を用いた数値実験において、最大8ノード×36ワーカーで良好な高速化を達成した。

    researchmap

  • 超巨大ニューラルネットの継続学習への型破りな線形代数技術の適用

    研究課題/領域番号:20K20624  2020年7月 - 2023年3月

    日本学術振興会  科学研究費助成事業 挑戦的研究(開拓)  挑戦的研究(開拓)

    横田 理央, Khan Emtiyaz, 大島 聡史, 伊田 明弘

      詳細を見る

    配分額:25350000円 ( 直接経費:19500000円 、 間接経費:5850000円 )

    最近の深層ニューラルネットワーク (DNN) の傾向を見ると、個々のタスクに特化した小規模なモデルを皆が冗長に学習するのではなく、大規模なモデルを用いて様々なタスクを一元的かつ継続的に学習する方向に向かっている。本研究では、このような超巨大ニューラルネットの継続学習を行う際に用いられるフィッシャー情報行列を高速に計算する方法を開発する。2021年度は、「分散並列二次最適化の BERT-Large への拡張」と「H 行列の GPU 実装と深層学習への応用」を計画当初の目標として掲げていたが、これらの目標は概ね達成できたといえる。分散並列二次最適化の BERT-Large への拡張に関しては、畳み込みニューラルネット(CNN)などよりも構造が単純なTransformerベースの BERT モデルは二次最適化に必要なフィッシャー行列のクロネッカー因子分解による近似が容易であり、BERT-Large への拡張は問題なく行うことができた。また、H 行列の GPU 実装に関しては、BLASのライブラリをcuBLASに置き換えることで高性能な GPU 実装が実現できた。H 行列では無数の小さな行列が生じるため batch 処理が必要になるが、これも MAGMA などのライブラリに標準的に実装されている機能であり、行列積だけでなく行列分解などの GPU 実装も行うことができた。深層学習への応用に関しては、2020年度に行った検証実験において、対角+低ランク行列によって作られるフィッシャー情報行列は階層的に分割した場合に非対角ブロックが低ランクになることが確認されており、この原理を応用することで大幅な高速化を実現することができた。クロネッカー因子分解ではO(N^1.5)の計算量になるが、H 行列分解はO(N)の計算量になることが実証できた。

    researchmap

  • Life-Long Deep Learning using Bayesian Principles

    研究課題/領域番号:20H04247  2020年4月 - 2023年3月

    日本学術振興会  科学研究費助成事業 基盤研究(B)  基盤研究(B)

    Khan Emtiyaz, Alquier Pierre, 横田 理央

      詳細を見る

    配分額:18200000円 ( 直接経費:14000000円 、 間接経費:4200000円 )

    researchmap

  • 機械学習向けハードウェアとの親和性が高い連立一次方程式の解法

    研究課題/領域番号:18H03248  2018年4月 - 2021年3月

    日本学術振興会  科学研究費助成事業 基盤研究(B)  基盤研究(B)

    横田 理央, 大島 聡史, 伊田 明弘

      詳細を見る

    配分額:16900000円 ( 直接経費:13000000円 、 間接経費:3900000円 )

    今年度の目標は低精度演算の有効活用であったが,これは概ね達成できたといえる.Tensor Coreを用いたQR分解の実装において16bitに起因する精度の低下を補正する手法を考案した.これは16bitの変数を2つ用いて仮数部を2倍に拡張することで情報落ちや丸め誤差を抑制するKahanの加算法に類似した手法である.ただし,Kahanの加算法がスカラーに対する手法なのに対して,本手法はそれを行列に拡張した手法となっている.この精度補正によりTensorCoreを用いたQR分解において単精度と同等の相対残差と直交性を実現することができた.また,このときの演算速度はTensorCoreの性能を最大限に引き出すことができていることを確認した.この成果は高性能計算分野のトップカンファレンスであるSC19のベストポスターの候補に選ばれた.
    このような低精度のQR分解を最大限に活用できる手法として階層的低ランク近似法が挙げられる.行列が低精度の浮動小数点で表される場合,それを低ランク近似することで精度をそれ以上劣化させることなく,演算量を大幅に低減できる.今年度は,階層的低ランク近似法として提案されているBLR,HODLR, HSS, H-matrix, H2-matrixの全ての手法を一般化した手法を提案し,C++の機能を最大限に活用することで簡素な実装を行うことができた.このフレームワークを用いて,これら5つの形式に対して行列の加算,行列ー行列積,LU分解,QR分解を容易に実装することができた.この成果は国内外の学会に採択され,現在ジャーナル論文も投稿中である.

    researchmap

  • H行列法ライブラリの機能拡張と次世代スパコン向け最適化

    研究課題/領域番号:17H01749  2017年4月 - 2020年3月

    日本学術振興会  科学研究費助成事業 基盤研究(B)  基盤研究(B)

    伊田 明弘, 横田 理央, 岩下 武史, 大島 聡史, 平石 拓

      詳細を見る

    配分額:18850000円 ( 直接経費:14500000円 、 間接経費:4350000円 )

    H行列法ライブラリHACApKの高機能化に関する研究を行った。H行列生成に動的負荷分散を導入し、H行列・ベクトル積計算に対してGPU向けアルゴリズムの開発・実装し、混合精度演算を導入した計算法を確立した。これらの新手法を用いた実装は、既存実装の数倍~十数倍の高速化を実現した。従来H行列に比べて効率的な演算・通信パターンを構築可能な新しい低ランク構造行列法である格子H行列法を提案した。数千プロセスを用いた実験では格子H行列・ベクトル積計算は既存実装の数十倍高速化を達成した。格子H行列に基づくLU分解法を開発し、BLR行列(単純な格子H行列)のQR分解法を開発した。これら分解手法の並列化も行った。

    researchmap

  • 社会インフラ映像処理のための高速・省資源深層学習アルゴリズム基盤

    2016年11月 - 2022年3月

    科学技術振興機構  CREST 

    篠田 浩一

      詳細を見る

    資金種別:競争的資金

    researchmap

  • 性能と生産性を両立するエクサスケールコンピュータ向け階層型粒子法フレームワーク

    研究課題/領域番号:16H02827  2016年4月 - 2019年3月

    日本学術振興会  科学研究費助成事業 基盤研究(B)  基盤研究(B)

    丸山 直也, 横田 理央, 田浦 健次朗

      詳細を見る

    配分額:16900000円 ( 直接経費:13000000円 、 間接経費:3900000円 )

    今日のペタスケールシステムの千倍の性能を目指したエクサスケールスーパーコンピュータでは、計算機アーキテクチャの質的および量的な変化が不可避であり、それに従って既存のアプリケーションの大幅な書き換えが必須となる。本研究では頻出基本数値計算手法である粒子法に着目し、アーキテクチャの変更の度にアプリケーションを変更することなく高性能を達成可能なソフトウェア基盤技術を確立することを目的として研究開発を進めた。これは、アーキテクチャ非依存にアプリケーション開発が可能なプログラミングフレームワークに基づき、並列化および性能最適化を自動化することを狙ったものである。今年度は本フレームワークの第一版としてCPUおよびGPUに対応したフレームワークを開発した。本フレームワークはC++テンプレートメタプログラミングに基づき、FMM等の階層的粒子法を簡便に記述可能なプログラミングモデルを提供する。ユーザプログラムはテンプレート展開によってCPU用の並列コードやCUDAを用いたGPU用コードへと自動的に変換されるため、対象プロセッサ用に別途プログラムを作成する必要がない。また、MPIを用いた複数ノード向け並列化もフレームワークによって自動的になされるため、単一のユーザプログラムによって単一プロセッサからスーパーコンピュータクラスの大規模システムまで統一的に動作させることが可能である。また、本フレームワークの実装には高性能を達成するためにFMMアルゴリズムの高性能実装技術に関する研究成果や軽量マルチスレッドランタイムであるMassiveThreadsが活用されており、人手による実装に近い性能が自動的に達成できている。

    researchmap

  • 基盤研究(B)

    2016年4月 - 2019年3月

    日本学術振興会  科学研究費助成事業 

    丸山 直也

      詳細を見る

    資金種別:競争的資金

    researchmap

  • FMMとH行列を組み合わせた大規模連立一次方程式の反復解法

    研究課題/領域番号:16H05859  2016年4月 - 2018年3月

    日本学術振興会  科学研究費助成事業 若手研究(A)  若手研究(A)

    横田 理央, リー シャオイエ, キース デイビッド

      詳細を見る

    配分額:6630000円 ( 直接経費:5100000円 、 間接経費:1530000円 )

    平成28年度にはFMMのH行列への拡張とH行列によるLU分解コードの開発を行った。この際、exaFMMで採用しているdual tree traversalをH行列のadmissibilityの判定に用いることでタスクベースの並列化を容易に行うことができた。平成29年度には内部カーネルのチューニングと実アプリケーションによるmultigrid法との比較を行った。Batched MAGMAを採用することにより小規模な行列をGPU上で高速に処理することができた。Multigrid法に対する優位性は行列の条件数に依存し、並列度がますとH行列が優位になることが分かった。

    researchmap

  • 若手研究(A)

    2016年4月 - 2018年3月

    日本学術振興会  科学研究費助成事業 

    横田 理央

      詳細を見る

    担当区分:研究代表者  資金種別:競争的資金

    researchmap

  • 研究活動スタート支援

    2015年8月 - 2017年3月

    日本学術振興会  科学研究費助成事業 

    横田 理央

      詳細を見る

    担当区分:研究代表者  資金種別:競争的資金

    researchmap

  • エクサスケーラブルな大規模連立一次方程式の前処理としてのFMMの代数学的拡張

    研究課題/領域番号:15H06196  2015年8月 - 2016年3月

    日本学術振興会  科学研究費助成事業 研究活動スタート支援  研究活動スタート支援

    横田 理央

      詳細を見る

    配分額:1430000円 ( 直接経費:1100000円 、 間接経費:330000円 )

    次世代計算機上で既に性能が出ると分かっている階層的 N 体アルゴリズムを出発点にとり,それを任意の連立一次方程式を扱えるソルバへと徐々に拡張した.平成27年度には,Poisson 方程式しか解くことのできない現在の FMM を Helmholtz 方程式や Stokes 方程式へと拡張し,流体解析のみならず構造・電磁場・音響解析へも適用できるようにした。また、それぞれの方程式を同等の計算条件,計算機環境の下で multigrid 法やHSS行列と直接比較し,いままでやられてこなかった手法間の定量的な優位性の評価を行った.Multigrid 法との比較においては Poisson 方程式に比べ Helmholtz 方程式は FMM の優位性が顕著であった.これは Helmholtz 方程式が高周波を含む場合に multigrid 法の収束性が著しく低下するのに対して,FMM の収束性がさほど低下しないことが原因である.HSS行列との比較ではセットアップのオーバーヘッドが小さい FMM が HSS に比べて合計の計算時間で有利になるという結果が得られた.特に2次元 Laplace 方程式においてその差は顕著で FMM が約1000倍高速であった.スケーラビリティのベンチマークにおいては FMM は Cray XC40 の 131,072 コアを用いた計算で良好な並列化効率が得られ,4000億点規模の計算を数秒で行うことができた.これは,FMM の計算としては世界最大規模であり,最速の計算でもあると思われる.

    researchmap

▼全件表示

担当経験のある科目(授業)

  • Computer Networks

     詳細を見る

  • 5類リテラシ

    機関名:東京工業大学 情報理工学院 情報工学系

     詳細を見る

  • 高性能科学技術計算

    機関名:東京工業大学 情報理工学院 情報工学コース

     詳細を見る

  • 計算機ネットワーク

    機関名:東京工業大学 情報理工学院 情報工学系

     詳細を見る

  • 5th Academic Group Literacy

     詳細を見る

  • High Performance Scientific Computing

     詳細を見る

▼全件表示