Funding entity Agencia Estatal de Investigación (AEI)
This research project addresses a foundational problem in the mechanistic modeling of complex systems and how so-called minimal models (surprisingly simple and universal descriptions) emerge. Faced with the exponential growth of data and the difficulty of identifying robust causal mechanisms, our proposal offers a unified framework that integrates large-deviation theory, information geometry, and non-equilibrium statistical physics, together with modern computational tools for Bayesian inference and machine learning. The expected scientific impact is twofold. On the one hand, we hope to establish new theoretical foundations that explain why very different systems converge towards the same families of effective models, providing a rigorous basis for scattered observations in the literature and solving a central problem in complexity science: the emergence of simplicity from heterogeneous interactions, noise, and multiscale dynamics. On the other hand, these advances translate into reproducible computational methodologies applicable to real-world problems where data are scarce or noisy, such as immune activation, the behavior of microbial communities, or the predictability of epidemiological dynamics. Thanks to collaboration with experimental groups and the team's expertise in immunology, microbial ecology, and quantitative epidemiology, the project generate tools to assess model robustness, quantify uncertainty, and rule out alternative hypotheses. This enhance the interpretation and design of experiments in disciplines where complexity has traditionally hindered mathematical formalization. From an internationalization perspective, the proposal aligns with global programs in simulation-based inference, physics-inspired machine learning, and probabilistic modeling in biosciences and socio-ecological systems, all of which are rapidly expanding. The results are published in high-impact open-access journals, and the code and notebooks are released on open platforms to facilitate adoption by the scientific community. Likewise, collaboration networks with leading European and American groups in statistical physics, mathematical biology, and probabilistic machine learning is strengthened. Finally, the project contribute to training young researchers in advanced modeling techniques, scientific computing, and uncertainty analysis, thereby consolidating national capabilities in a key area for future interdisciplinary research.
Layman's summary: This project blends statistical physics and machine learning to explain why complex systems follow simple, universal models. It provide computational tools to analyze noisy data across immunology, microbial ecology, and epidemiology.
Techniques employed: bayesian inference, naural networks, Probabilistic AI, causal methods
MINIMODELS
