absorb.md

Mitsuhisa Sato

Chronological feed of everything captured from Mitsuhisa Sato.

Large-Scale Hybrid Quantum-Classical Electronic Structure Calculations

This paper demonstrates a closed-loop workflow integrating a Heron quantum processor with the Fugaku supercomputer (152,064 nodes) for electronic structure calculations. This hybrid approach enables approximations of chemistry models beyond exact diagonalization, achieving accuracy comparable to classical approximation methods. The work highlights advancements in orchestrating computational resources at an unprecedented scale for current supercomputers.

Parallelization Strategies for IPOP-CMA-ES Achieve Superlinear Speedups

This paper details methods for accelerating the IPOP-CMA-ES algorithm on high-performance computing architectures. By integrating BLAS/LAPACK routines and employing two distinct parallelization strategies—sequential-ordering and concurrent processing—the researchers achieved substantial, including superlinear, speedups on a 6144-core supercomputer. The concurrent processing strategy demonstrated superior performance.

Fugaku Supercomputer Accelerates MRI Data Processing with FSL

This preliminary study demonstrates the successful application of the Fugaku supercomputer for accelerating MRI data processing using the FMRIB Software Library (FSL). The study found high consistency in tensor-based measurements and subcortical structure segmentations between Fugaku and conventional systems, while significantly reducing processing time. This indicates a viable solution for handling the growing volume of MRI data and the computational demands of its analysis.

CORTEX: A Novel Brain Simulator for Large-Scale Neuromorphic Computing on Fugaku

CORTEX is a new algorithmic framework for large-scale brain simulation, leveraging the Fugaku Supercomputer. Its core innovation, Indegree Sub-Graph Decomposition, enables efficient parallel processing of synaptic interactions by segmenting the global graph into independent sub-graphs. This approach eliminates data racing without mutexes and enhances communication-computation overlap, leading to significant performance improvements over existing solutions like NEST Simulator.