Пленарная секция

Monday, September 28, 2026, 10:00-11:15
Keynote Session 1
Conference Hall

Moderator: Chetverushkin Boris Nikolaevich

Simultaneous interpretation between Russian and English will be provided during the plenary session. Interpretation will be available via the online stream links found on the registration page.If you plan to use the interpretation service, please bring headphones compatible with your electronic device.

Conference Opening
Chetverushkin Boris Nikolaevich, Chair of the Conference Program Committee, Scientific Director of the Keldysh Institute of Applied Mathematics of the Russian Academy of Sciences, Academician, Doctor of Physical and Mathematical Sciences, Professor

Diffuse Boundary Models: "Digital Core" and Other Applications
V.A. Balashov, E.V. Zipunova, A.S. Ponomarev,
Savenkov Evgeniy Borisovich, Doctor of Physical and Mathematical Sciences, Corresponding Member of the Russian Academy of Sciences, Deputy Director for Research at the Keldysh Institute of Applied Mathematics of the Russian Academy of Sciences

The report examines various aspects of implementing the "Digital Core" applied technology, which is based on the direct numerical simulation of multiphase fluid micro-flows within the pore space of oil and gas reservoir rocks. The technology relies on so-called diffuse-interface models. These models are thermodynamically consistent and enable the description of multiphase fluid flows by directly resolving phase interface dynamics and contact angles using a spatially uniform approach. Diffuse-interface models belong to the class of weakly nonlocal—or gradient—thermodynamic models, allowing for the description of a wide range of processes involving the evolution of interfaces. A key feature of these models is that thermodynamic potentials depend not only on the model's state variables but also on their gradients. The equations employed in this work are based on density gradient theory and represent a variant of the Navier-Stokes–Cahn-Hilliard system. The developed computational algorithms allow for efficient parallelization and enable the analysis of problems involving significant grid dimensions—including flows within voxel-based models of real porous media obtained via micro-computed tomography. The report discusses the theoretical and practical aspects of the technology, the specifics of the computational algorithms, and outstanding theoretical and applied challenges requiring resolution. In addition to hydrodynamic applications, the report considers other uses for this class of models, specifically the simulation of electrical breakdown channel propagation in micro-inhomogeneous media.

Mixed-precision Programming Environment on Tianhe System
Su Xing, Associate Professor, College of Computer, National University of Defense Technology (NUDT), China

In the post-Moore's Law era, exascale computing faces two major challenges: the storage wall and the energy wall. Driven by AI computing, low-precision computing power is rapidly growing in mainstream computing chips, including CPUs, GPUs, and other accelerators. Mixed-precision computing emerges as a promising approach to improving performance and reducing energy consumption by replacing some high-precision computations with low-precision ones. We have developed a comprehensive programming environment for mixed-precision computing on Tianhe supercomputers. This comprises a compiler, precision analysis tools and mixed-precision algorithm libraries. Performance validation has been conducted on multiple scientific computing applications involving tens of millions of cores.

Chronicles of the Emergence of AGI
Konushin Anton Sergeyevich, Candidate of Physical and Mathematical Sciences, Head of the Spatial Intelligence Research Group at the AIRI Institute, Associate Professor at the Faculty of Computational Mathematics and Cybernetics (Lomonosov Moscow State University) and the Higher School of Economics (HSE University), and Program Director of the Intellect Foundation.

Monday, September 28, 2026, 11:45-13:30
Keynote Session 2
Conference Hall

Moderator: Chetverushkin Boris Nikolaevich

Full-wave seismic modeling using the spectral element method and its massively parallel implementation on the MSU-270 supercomputer
Vershinin Anatoly Viktorovich, Faculty of Mechanics and Mathematics, Lomonosov Moscow State University; Fidesys
Yu.P. Ampilov, A.M. Antonov, V.A. Levin

This report examines the formulation and solution of a 3D dynamic elasticity problem within the context of full-scale seismic survey modeling, using an oil field in Western Siberia (covering approximately 200 square kilometers) as a case study. The solution algorithm is based on the spectral element method (SEM) and an explicit time-integration scheme; it was developed for the numerical simulation of wave propagation in heterogeneous 3D media characterized by rapidly varying rock properties. The report discusses the technical aspects of implementing this algorithm on a massively parallel multi-GPU system using CUDA technology. The developed algorithm was verified against the analytical solution to Lamb's problem, and the p-convergence of the simulation results was analyzed for various SEM orders. The parallelization efficiency of the SEM was evaluated on the MSU-270 hybrid supercomputer across different spatial discretization orders and time-integration scheme parameters. Over the course of two months of computations on the MSU-270 supercomputer, a full-wave simulation of the seismic field was performed—involving 12,285 sources and a detailed geophysical model comprising approximately 5.6 million cells and nearly 3 billion SEM nodes (totaling about 9 billion degrees of freedom). A comprehensive wavefield was obtained for each source, capturing surface Rayleigh waves as well as the full spectrum of body, reflected, refracted, and diving waves, including longitudinal, transverse, and converted waves. All forms of diffraction and multiple reflections were accounted for—capturing the full range of phenomena occurring in a real geological environment during seismic exploration. This comprehensive wavefield modeling is relevant for exploring the capabilities of modern seismic data processing and interpretation methods, which currently rely on simplified assumptions regarding subsurface structure. We anticipate that the developed technology will be widely adopted in routine seismic exploration practice.

Federated Learning and Supercomputing: New Opportunities for International Cooperation
Fabio Borges de Oliveira, director and a professor at the National Laboratory for Scientific Computing (LNCC), Brazil

Federated Learning enables institutions to collaboratively train AI models while keeping data distributed and under local control. Combined with modern supercomputing infrastructures, it creates new possibilities for cooperation among HPC centers, universities, research institutions, and national scientific infrastructures. This talk explores how federated learning and supercomputing can provide a technological foundation for deeper scientific cooperation. Starting from the relationships developed through the BRICS HPC and AI community, we discuss opportunities for connecting distributed computational resources and datasets while preserving institutional control, security, and privacy. Potential areas of collaboration include AI for science, large-scale and multimodal models, federated training, secure aggregation, and privacy-preserving technologies. In the spirit of the scientific cooperation strongly promoted by Professor Vladimir Voevodin, the talk aims to identify concrete opportunities for joint research and a distributed ecosystem for AI and high-performance computing.

Moscow University’s Supercomputing Complex: From General-Purpose Computers to Implementations of Co-Design Principles
Nikitenko Dmitry Alexandrovich, Head of the Scientific and Technical Department, Leading Researcher at the Research Computing Center of Lomonosov Moscow State University, Candidate of Physical and Mathematical Sciences.

Insights from an IT Systems Integrator: How We Built the Infrastructure for Agents Within SENSE
Zhelvis Vladimir Algirdovich, Deputy Head of the Internal Development Department at SENSE

T-Bank R&D Center: research into distributed systems and other industry challenges
Teplov Alexey Mikhailovich, Head of Databases and Analytic Engines R&D at the T-Bank R&D Center, Head of the T Project and Training Laboratory at HSE University, Candidate of Physical and Mathematical Sciences

Tuesday, September 29, 2026 г., 17:00-17:30
Keynote Session 3
Conference Hall

Moderator: Chetverushkin Boris Nikolaevich

Awarding and Conference Closing
Chetverushkin Boris Nikolaevich,
Chair of the Conference Program Committee, Scientific Director of the Keldysh Institute of Applied Mathematics of the Russian Academy of Sciences, Academician, Doctor of Physical and Mathematical Sciences, Professor