Prof. Witold Dzwinel

AGH University of Krakow, Poland



Title of the paper: Fast Simulations and Surrogate Models of Complex Systems. Modeling the Dynamics of a Tumor Under Therapy


Abstract: Predicting the spatio-temporal evolution of highly complex systems- particularly in medical and biological contexts - remains one of the grand challenges of modern science. Traditional computational models, while powerful, are often over-parameterized, computationally intensive, and limited in their ability to represent dynamic processes unfolding across multiple spatial and temporal scales. In oncology, for instance, cyclic anticancer therapies continually reshape tumor heterogeneity and remodel the surrounding microenvironment, making parameters derived from historical data quickly outdated and restricting the clinical usefulness of purely mechanistic cancer models for guiding personalized treatment. In this plenary talk, I will present and discuss concepts and methodological frameworks from our recent research project aimed at overcoming these limitations by developing novel, fast, and robust modeling strategies integrated with artificial intelligence. Our goal is to develop modeling strategies that deliver reliable forecasts under uncertainty, operate efficiently, adapt flexibly to new data, remain robust when conditions change, and - most importantly - offer interpretable insights that can guide real scientific and clinical decisions. Within this context, we explore three complementary frameworks: (1) theory-driven computer supermodeling, (2) unsupervised and physics-inspired machine learning (PIML), and (3) large language models (LLMs). The latter are envisioned as retrieval-augmented generation (RAG) systems - acting as knowledge integrators and natural language interfaces - to complex simulations. The main scientific contributions include:
• Development of efficient simulation methodologies based on ensemble supermodeling, knowledge distillation, and hybrid PIML approaches.
• Design of a RAG framework with domain-specific integration for enhanced adaptability.
• Coupling of RAG/LLM architectures with simulation surrogates, enabling a novel paradigm of language-driven therapy design.
The talk will highlight two key research directions. The first is the creation of fast, uncertainty-aware simulation frameworks for complex dynamical systems. The second is their application to predictive oncology, where we envision embedding simulation modules within RAG-enhanced LLMs. Together, these contributions not only advance the state of the art in computational modeling and clinical decision support but also lay the groundwork for interpretable, expert-guided AI systems that bridge scientific computing and precision medicine.
“Research project supported by the program ‘Excellence Initiative – Research University’ at AGH University of Krakow.”

Bio: Dr. Witold Dzwinel is a Full Professor at the Faculty of Computer Science, AGH University of Krakow, where he leads the Complex Systems Group. His research focuses on applying diverse modeling and simulation paradigms to the study of complex systems, ranging from discrete and continuous formalisms to recent advances in physics-based machine learning (PBML). Simultaneously, he develops methods for interactive visualization of multidimensional data, such as the IVHD algorithm, by exploiting hybridized modeling approaches. From 1988 to 1992, he worked at the Joint Institute for Nuclear Research in Dubna as a member of IBR-2 Nuclear Safety research group, where he began applying machine learning methods to develop the IBR-2 surveillance system. Between 1999 and 2002, he was a Research Scholar at the Minnesota Supercomputing Institute, University of Minnesota, continuing his pioneering work on interacting particle paradigms - such as Molecular Dynamics, Dissipative Particle Dynamics, the Fluid Particle Method, and Smoothed Particle Hydrodynamics - for simulating complex fluids across various supercomputer architectures. In parallel, he used machine learning to analyze simulation data, ranging from the detection of liquid crystals and aggregates in complex fluids to earthquake modeling. After returning to AGH, Professor dr Witold Dzwinel extended his research to hybrid models combining interacting particle approaches with cellular automata, known as the Particle Automata Model (PAM). This innovative framework bridges microscopic particle-based dynamics with rule-based automata, allowing the study of emergent behaviors across multiple scales. PAM has been successfully applied to diverse domains including infection spread, blood dynamics in capillaries, Fusarium Graminearum infection in wheat, and cancer dynamics. His current research explores the integration of supermodeling and machine learning for developing surrogate simulations of complex phenomena, with applications ranging from tumor dynamics under anti-cancer therapy to heavy ion collisions in CERN’s ALICE experiment. He is the author of about 200 scientific papers, primarily in the field of computational science. https://www.researchgate.net/profile/Witold-Dzwinel/research




Categories