Distributed by design
We study learning when data cannot or should not be centralized: heterogeneous clients, non-IID data, privacy constraints and limited communication.
Research at M.O.D.A.L.
AI that learns across data, physics and systems
We develop Artificial Intelligence methods for settings where data are distributed, systems are heterogeneous, physical knowledge matters and decisions must remain reliable under real-world constraints.
Distributed by design
Physics-informed
Resource-aware
Application-grounded
M.O.D.A.L.methods × systems× impact
Intelligence Trustworthy
AI Scientific
ML Generative &
Agentic AI Edge · Cloud
· HPC Digital
Twins Healthcare Mobility Industry Science
Why M.O.D.A.L. is distinctive
Methodological research,
built for non-ideal worlds
Our work starts from methodological questions, then follows them through algorithms, computing systems and demanding application domains.
Science inside the model
Mathematical structure, governing equations and domain knowledge are not afterthoughts: they become part of how models learn and reason.
System-aware intelligence
Algorithms are designed together with their computational reality, from constrained edge devices to distributed multi-GPU and HPC environments.
Transfer across domains
We look for methods that remain meaningful beyond one dataset: from healthcare and mobility to industry, geoscience, IoT and scientific computing.
Interactive research map
Six pillars. One connected agenda
Select a pillar to see how methods connect to application domains and to the most recent research output.
01
Distributed intelligence
Federated & Distributed Intelligence
FL
We investigate federated and decentralized learning under realistic heterogeneity: non-IID data, diverse devices, limited communication, personalization, one-shot learning and continual adaptation.
Where it connects
Research now
Research pillars
Methods that reinforce each other
The boundaries are intentionally porous: many of our strongest results sit at the intersection of two or more pillars.
01
Federated & Distributed Intelligence
Aggregation, non-IID learning, one-shot FL, personalization, continual learning and distributed intelligence across heterogeneous nodes.
Deep dive: Federated Learning →✓
02Trustworthy & Responsible AI
Privacy, security, robustness, explainability and machine unlearning for AI systems that must remain accountable and controllable.
∂u/∂t+𝒩[u]=0
03Scientific Machine Learning
Physics-informed neural networks, hybrid models, inverse problems and learning strategies that encode mathematical and physical structure.
Deep dive: Scientific ML →04
Generative & Agentic AI
Synthetic data, generative models, reasoning pipelines, multi-agent systems and knowledge-centric AI for complex data and decision workflows.
05
Efficient AI & Edge–Cloud–HPC
Lightweight models, communication efficiency, energy-aware learning, parallel algorithms and scalable execution from edge to multi-GPU systems.
PHYDIG
06Digital Twins & Adaptive Systems
Data-driven and model-based digital representations that support prediction, what-if analysis, monitoring and adaptation over time.
Featured · Nature Communications 2025A digital twin framework for urban parking management and mobility forecastingOpen paper ↗From methods to impact
Applications are testbeds,
not silos
We use demanding domains to expose methodological weaknesses and to validate whether new AI approaches remain useful outside ideal benchmark settings.
Healthcare & Precision Medicine
Privacy-sensitive learning, multimodal data, medical IoT, clinical forecasting and decision support.
Federated AITrustworthy AIGenerative AI
Smart Cities & Mobility
Traffic and parking prediction, connected mobility, urban sensing, autonomous navigation and what-if analysis.
Digital TwinsDistributed AIGenerative AI
Smart Industry & Maintenance
Predictive maintenance, industrial IoT, production intelligence and resilient cyber-physical systems.
Edge AIDigital TwinsTrustworthy AI
Energy & Sustainability
Energy-aware learning, smart energy systems, efficient computation and sustainability-aware AI design.
Efficient AIEdge–HPCGenerative AI
Geoscience & Environment
Seismology, Earth observation, environmental modelling and scientific inference over complex spatial data.
SciMLHPCDeep Learning
Autonomous Systems & IoT
Connected devices, sensor intelligence, autonomous navigation and learning across the computing continuum.
Edge AIFederated AIAgentic AI
Cybersecurity
Privacy-preserving analytics, anomaly detection, secure distributed learning and robust data generation.
Trustworthy AIGenerative AIFederated AI
Scientific & Engineering Systems
Inverse problems, PDE-driven learning, scalable simulation and hybrid models for physical systems.
SciMLHPCDigital Twins
Research now
Recent directions from the publication catalogue
Loading recent research…
Deep dives
Go deeper into two core research lines
01Explore →
02Explore →
Federated Learning
Non-IID data, aggregation, energy efficiency, HPC environments, TinyFL, one-shot learning and the evolution toward trustworthy distributed intelligence.
Scientific Machine Learning
Physics-Informed Neural Networks, free-boundary problems, option pricing, groundwater flow, monitoring and emerging directions in physics-informed learning.