Structured Machine Learning Lab
We work on Machine Learning for domains with complex structure like sequences, trees, graphs and relational knowledge, combining statistical and symbolic approaches with a strong focus on human-centric AI: neuro-symbolic integration, explainable & interactive ML, and constructive preference elicitation designed to support, not replace, human decision-making.
Research areas
Neuro-Symbolic AI
Integrating logic and constraint reasoning with deep learning, and studying failure modes like reasoning shortcuts in concept-based models.
Structured Prediction
Learning to predict and optimize over sequences, trees, graphs and combinatorial structures.
Explainable & Interactive ML
Interpretability of large models, human-machine decision making, and algorithmic recourse.
Preference Elicitation
Constructive preference elicitation for recommendation and human-centered decision support.
Uncertainty & Calibration
Calibrated confidence estimates and learning-to-defer frameworks for reliable human-AI collaborative decision making.
Geometric Deep Learning
Explainability, self-explainability, temporal dynamics and neuro-symbolic integration for Graph Neural Networks.
Projects
TANGO
Synergistic human-machine decision making — hybrid decision support systems that align humans and machines in values and goals.
TAILOR Network
European network on Trustworthy AI, connecting research groups working on foundations of human-centric, reliable AI.
FAIR
Future Artificial Intelligence Research — the Italian national foundation on AI, of which the lab is an active research unit.