High-Performance Computing × Generative AI
片桐 孝洋 / Professor, Nagoya University
Professor, Data Science Research Division
Information Technology Center, Nagoya University
Concurrent appointment: Graduate School of Informatics / School of Informatics, Nagoya University
My research centers on high-performance computing (HPC), with a particular focus on generative AI and AI agents for HPC code generation, software auto-tuning, GPU and massively parallel computing, numerical linear algebra, large-scale AI, and hybrid quantum-classical computing. A central goal is to move HPC software development from a model in which humans write and tune everything manually toward one in which humans and AI collaboratively design, generate, verify, evaluate, and optimize high-performance software.
Information Technology Center, Nagoya University
News & Updates
Selected recent developments in HPC-GENIE, agentic HPC, research outputs, and international activities.
Research
My work connects numerical algorithms, high-performance software, machine learning, and advanced computing systems.
LLM-based generation, debugging, verification, and performance optimization of C, Fortran, CUDA, and other HPC codes. Topics include multi-agent orchestration, local LLMs, knowledge accumulation, autonomous experimentation, and agentic performance engineering.
Automatic selection and adaptation of algorithms, implementations, and parameters according to architectures, inputs, and performance objectives. Research topics include auto-tuning languages, performance modeling, search, mixed precision, and run-time optimization.
Auto-Tuning Engineering →High-performance implementations for GPUs, multicore and many-core processors, clusters, and supercomputers. Technologies include CUDA, OpenMP, OpenACC, MPI, memory optimization, communication reduction, and performance portability.
Large-scale eigenvalue problems, LU/QR factorizations, Krylov subspace methods, GMRES, BiCGStab, preconditioning, mixed-precision algorithms, verified numerical computation, and parallel/GPU implementations.
Parallel eigensolver archive →Large-scale and distributed machine learning, explainable AI, acceleration of machine-learning workloads, and the use of AI techniques for numerical computing and performance optimization.
Hybrid quantum-classical computing, quantum and quantum-inspired optimization, HPC acceleration of quantum algorithms, and auto-tuning for heterogeneous quantum-classical workflows.
Current Projects
Current research platforms at the intersection of generative AI, agentic systems, and high-performance computing.
High-Performance Computing with Generative Neural Intelligence for Execution. HPC-GENIE studies generative-AI technologies for HPC code generation, automatic optimization, RAG and fine-tuning, local LLMs, mixed precision, accuracy assurance, explainable AI, and software auto-tuning.
Agentic Computing Research Unit. ACRU studies multi-agent systems that integrate requirements analysis, code generation, compilation, execution, performance measurement, performance analysis, auto-tuning, debugging, and knowledge accumulation for HPC research and software development.
A CLI-oriented multi-agent framework in which multiple AI coding agents collaborate on HPC code generation, compilation, execution, performance measurement, and iterative optimization. It is a core open-source research output of the HPC-GENIE / ACRU effort.
Academic Profile
Professor Takahiro Katagiri is a computer scientist specializing in high-performance computing, parallel numerical computation, and software auto-tuning. Since the late 1990s, he has worked on auto-tuning technologies for performance portability through the development of parallel numerical libraries such as ILIB and ABCLib.
His research has addressed automatic selection and adaptation of algorithms, implementations, block sizes, and parallelization strategies according to processors, memory hierarchies, communication mechanisms, compilers, and input characteristics. He proposed the FIBER framework, which integrates optimization at installation time, before execution, and during run time, and developed ABCLibScript for describing expert optimization knowledge and connecting it to code generation, performance modeling, and parameter search.
His recent work extends these ideas toward generative AI and agentic systems for HPC. The goal is to automate not only parameter tuning, but also requirements analysis, code generation, compilation, execution, performance evaluation, debugging, iterative improvement, and knowledge accumulation on real computing systems.
Publications & Research Outputs
For the most complete and current publication record, please use the official Nagoya University faculty profile or Google Scholar. The local publication archive is organized by Japanese academic year (April–March).
For Prospective Students
We conduct research across high-performance computing, generative AI, AI agents, GPU computing, numerical algorithms, large-scale machine learning, quantum-related computing, and practical research software.
Our research is centered on high-performance computing and the effective use of cutting-edge supercomputers. Topics include massively parallel computing, large-scale AI, generative AI and AI agents, quantum computing technologies, GPU computing, CPU/GPU performance optimization, numerical algorithms, and practical software tools. We are also open to new research topics when they have a clear connection to HPC and computational science.
Applicants from Nagoya University and other universities may apply to the Department of Computing and Software Systems, Graduate School of Informatics.
Doctoral applicants should contact Professor Katagiri well before application to discuss the proposed research topic, prior research achievements, and technical background.
Research on LLMs and AI agents in this laboratory focuses primarily on program generation and optimization for supercomputers and HPC systems.
Researchers interested in HPC, generative AI, agentic systems, software auto-tuning, numerical computing, GPU computing, or quantum-HPC integration are welcome to contact Professor Katagiri regarding possible opportunities.
Books & Education
Textbooks and technical books on parallel programming, numerical computing, HPC, and software auto-tuning.
Awards & Academic Service
Collaboration
I welcome discussions on collaborative research in HPC, generative AI, numerical computing, GPU computing, performance engineering, and quantum-HPC integration.
Please contact me about joint research, supercomputer use, parallelization, numerical algorithms, performance optimization, and AI-assisted scientific software development.
Collaborative research using Nagoya University's institutional frameworks and computing resources may be possible, subject to available personnel, institutional review, applicable laws and regulations, and resource-use policies.
Consultation is possible for seminars and training on HPC, parallel programming, GPU computing, performance optimization, and generative-AI-assisted HPC code development.
Archive
Older pages are retained for historical reference. For current information, please use the official faculty profile, the laboratory site, and Google Scholar.
Contact
For research collaboration, invited talks, student applications, and other academic inquiries, please contact me by email. Prospective students should include their current affiliation, intended degree program, research interests, and relevant programming/HPC experience.
Email: katagiri "at" cc.nagoya-u.ac.jp
Replace "at" with @.