Assistant Professor (RTD/a) at the University of Milano-Bicocca (DISCo), in the Intelligent Sensing Laboratory. Ph.D. in Mathematics. I work where formal methods meet machine learning: logic-based and neuro-symbolic representations for sequential and multimodal data, and explanations that can be inspected, validated, and monitored rather than merely plotted.
The through-line of everything below is one commitment: a system should be able to say how it reached a conclusion, and refuse to claim more than its evidence supports.
Interval temporal logic and symbolic learning. My doctoral and post-doctoral work is in modal and interval temporal logics (Halpern–Shoham and its decidable fragments) and in modal symbolic learning — decision trees, rule extraction, and classifiers whose models are logical formulas over intervals rather than opaque functions over feature vectors. The point is not interpretability as decoration: a logical model is an object you can check, refine, and monitor.
Neuro-symbolic AI, and its failure modes. Combining learned perception with symbolic reasoning is only an advance if the symbolic layer actually constrains the answer. Much of my recent work is diagnostic: showing where neuro-symbolic systems reach the right answer for the wrong reason, and building the instruments that detect it.
Formal explainable AI (FXAI) and conformance. Explanations that carry a guarantee — derived from the model's own inference rather than approximated beside it — and, increasingly, the question of what evidence a legal duty actually requires from a decision system.
Applied setting. Healthcare and telemonitoring above all, through the ANTHEM national programme (PNC): physiological time series and multimodal signals, where an unexplainable model is not deployable. Previously iNEST (PNRR) on time-series analytics, and industrial work on rare-event prediction and rule extraction for gas-turbine trip prediction with Siemens Energy.
reasonsmith — Statute → formal duty → graded evidence → refusal-honest report. It asks what a decision system can establish about a legal requirement, what its evidence can only witness, and what the report must refuse to pretend it knows. Evidence is graded on an explicit lattice, and a duty that cannot be discharged honestly is never reported as satisfied. It can re-run a system's own inference to identify which legally-owed reasons a decision actually depended on — and which were deleted. Open to contributors: packs (regulations), engines (verifiers), and systems under test.
nesyarena — Accuracy Hides Semantic Error: a diagnostic arena for neuro-symbolic reasoning. Systems whose decisions disagree with the semantics they claim can still score well on accuracy. This measures that gap instead of assuming it away.
ledgerpress — A single-source academic record published as both a website and a printed CV, with a build that refuses to publish a contradiction. The consistency checks are the product. Powers eduardstan.github.io.
Sole.jl — The Julia ecosystem for symbolic learning and interval-temporal reasoning, developed at the ACLAI Lab, where I am an organization owner and helped originate the framework. Components including SoleLogics.jl (modal and interval logics, Allen relations, finite model checking) and SoleReasoners.jl (many-valued multi-modal tableaux, including Halpern–Shoham) implement results from my research on interval temporal logic and modal reasoning.
CouncilAgent — Multi-LLM council: fan-out, deliberate, aggregate, calibrated confidence. Pluggable topologies and aggregation strategies over a pure asyncio core.
Associate Editor — Neurocomputing (Elsevier, IF 6.5, Q1) and Frontiers in Artificial Intelligence (Pattern Recognition). Lead Editor of the special issue on Explainable AI for Trustworthy and Human-Centric Healthcare.
Area Chair — IJCNN (2025, 2026), ICASSP (2026). Program Committee — AAAI, ICLR, IJCAI, ECAI. Technical Committee — IEEE Consumer Technology Society (MDA). Tutorials/Special Session Co-Chair — IEEE ICIR 2026.
Teaching — Databases (B.Sc.), Intelligent Consumer Technologies (M.Sc. AI for Society and Technology), Foundations of AI (2nd-level Master, DAI4Health). 10+ B.Sc. and 5+ M.Sc. theses supervised.
Website · Google Scholar · ORCID · LinkedIn
Open to collaboration on trustworthy, interpretable, and neuro-symbolic AI — particularly where a system has to justify itself to someone who is entitled to an answer.