High Performance Artificial Intelligence Systems Research Team

Research summary

The High Performance Artificial Intelligence Systems Research Team is an R-CCS laboratory focusing on convergence of HPC and AI, namely high performance systems, software, and algorithms research for artificial intelligence/machine learning. In collaboration with other research institutes in HPC and AI-related research in Japan as well as globally, it seeks to develop next-generation AI technology that will utilize state-of-the-art high-performance computation facilities, including Fugaku. Specifically, we conduct research on next-generation AI systems by focusing on the following topics:

Extreme speedup and scalability of deep learning:
Achieve extreme scalability of deep learning in large-scale supercomputing environments including the post-K, extending the latest algorithms and frameworks for deep learning.
Performance analysis of deep learning:
Accelerate computational kernels for AI over the state-of-the-art hardware architectures by analyzing algorithms for deep learning and other machine learning/AI, measuring their performance and constructing their performance models.
Acceleration of modern AI algorithms:
Accelerate advanced AI algorithms, such as ultra-deep neural networks and high-resolution GAN over images, those that require massive computational resources, using extreme-scale deep learning systems.
Acceleration of HPC algorithms using machine learning:
Accelerate HPC algorithms and applications using empirical models based on machine learning.
Intelligent programming systems:
Use AI to auto-generate programs that can adapt to and withstand the complexity and divergence of hardware design.

Representative papers

Jintao Meng, Chen Zhuang, Peng Chen, Mohamed Wahib, Bertil Schmidt, Xiao Wang, Haidong Lan, Dou Wu, Minwen Deng, Yanjie Wei, Shenzhong Feng.:
"Automatic Generation of High-Performance Convolution Kernels on ARM CPUs for Deep Learning",
IEEE Transactions on Parallel & Distributed Systems, vol. 34, April 2022.
Jintao Meng, Peng Chen, Mingjun Yang, Mohamed Wahib, Yanjie Wei, Shengzhong Feng, Wei Liu, Junzhou Huang.:
“Boosting the Predictive Performance with Aqueous Solubility Dataset Curation”,
Nature Scientific Data, March 2022.
Truong Thao Nguyen, Francois Trahay, Jens Domke, Aleksandr Drozd, Emil Vatai, Jianwei Liao, Mohamed Wahib, Balazs Gerofi.:
"Why Globally Re-shuffle? Revisiting Data Shuffling in Large Scale Deep Learning",
36th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2022).
Albert Khaira, Truong Thao Nguyen, Leonardo Bautista Gomez, Ryousei Takano, Rosa Badia, Mohamed Wahib.:
"An Oracle for Guiding Large-Scale Model/Hybrid Parallel Training of Convolutional Neural Networks",
30th ACM International Symposium on High-Performance Parallel and Distributed Computing (HPDC 2021).
Peng Chen, Mohamed Wahib, Xiao Wang, shinichiro takizawa, Takahiro Hirofuchi, Ogawa Hirotaka, Satoshi Matsuoka.:
"Performance Portable Back-projection Algorithms on CPUs: Agnostic Data Locality and Vectorization Optimizations",
35th ACM International Conference on Supercomputing (ICS 2021).
Peng Chen, Mohamed Wahib, Xiao Wang, Takahiro Hirofuchi, Hirotaka Ogawa, Ander Biguri, Richard Boardman, Thomas Blumensath, Satoshi Matsuoka.:
"Scalable FBP Decomposition for Cone-Beam CT Reconstruction",
International Conference for High Performance Computing, Networking, Storage, and Analysis (SC 2021).
Fareed Mohammad Qararyah, Mohamed Wahib, Doga Dikbayır, Mehmet Esat Belviranl, Didem Unat.:
“A computational-graph Partitioning Method for Training Memory-constrained DNNs”,
Elsevier Parallel Computing, Volume 104 pp. 102-117, July 2021.
Jens Domke, Emil Vatai, Aleksandr Drozd, Peng Chen, Yosuke Oyama, Lingqi Zhang, Shweta Salaria, Daichi Mukunoki, Artur Podobas, Mohamed Wahib, Satoshi Matsuoka.:
"Matrix Engines for HPC: A Performance Study from the Applications Perspective",
35th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2021).
Mohamed Wahib, Haoyu Zhang, Truong Thao Nguyen, Aleksandr Drozd, Jens Domke, Lingqi Zhang, Ryousei Takano, Satoshi Matsuoka.:
"Scaling Deep Learning Workloads Beyond Memory Capacity",
International Conference for High Performance Computing, Networking, Storage, and Analysis (SC 2020).
Chen Peng,Wahib Mohamed,Takizawa Shinichiro,Matsuoka Satoshi.:
"A Versatile Software Systolic Execution Model for GPU Memory Bound Kernels",
International Conference for High Performance Computing, Networking, Storage, and Analysis (SC 2019).

Team Leader Mohamed Wahib

Keyword