Models that learn from data while respecting the differential equations that describe physical, chemical and physiological processes.
Research group at IDSIA USI-SUPSI, Lugano, Switzerland
Knowledge-Informed AI
The Knowledge-Informed AI group (KIAI) develops machine learning methods that combine data with what is already known about a system: physical and chemical laws, cause-and-effect relations and expert knowledge. The aim is models that remain accurate, interpretable and reliable when data are scarce, heterogeneous or distributed across institutions.
The methods are developed and validated together with hospitals, companies and research institutions, with applications in healthcare, life sciences, industry and materials.
Research lines
Knowledge enters our models through their structure (differential equations, causal graphs), through their prior distribution (hierarchical models that encode how data sets are related) or through their training objective. Each research line has a page with our methods, projects, publications and software.
Causal models that estimate the effect of interventions and answer what-if questions, built on probabilistic graphical models that combine data with expert knowledge.
Coordinated by Daniele Malpetti
Methods and software to train models on data that remain at the institutions that collected them, such as hospitals and research centres.
Application domains
The same methodological core serves very different problems. These are the domains in which our methods have been applied.
-
Healthcare
Decision support for prevention, diagnosis and treatment, developed and validated with hospitals in Switzerland and Italy.
Projects in this domain -
Life sciences
Interpretable models for transcriptomics and single-cell data, and privacy-preserving analysis of multi-centre biomedical data.
Projects in this domain -
Industry and engineering
Models of materials and production processes that combine experimental data with physical and chemical knowledge, and data-driven decision support for companies.
Projects in this domain -
Environment and society
Risk mapping and surveillance tools for public health and natural hazards, and analytics for public institutions.
Projects in this domain
News
- 2026
-
2026
Laura Azzimonti serves as Guest Editor of the npj Digital Medicine collection "Physics-Informed Machine Learning for Personalized Disease Prediction, Prevention, and Management".
- 2026
-
2026
Laura Azzimonti gives an online lecture on physics-informed machine learning and causal AI for personalised diabetes prevention in the Causal Analysis of Biomedical Data Lecture Series of the Luxembourg Centre for Systems Biomedicine, University of Luxembourg.