Optimization at the University of Pisa, coming from symbolic AI and working between the two. What holds them together is one idea: the difficulty of a problem lies in how it is written, so the useful move is to rewrite it, as a set of smaller problems or in a different formalism altogether. Knowledge representation and expert systems on one side, decomposition methods for large-scale optimization on the other, and learned heuristics inside exact solvers where the two meet.
→ dmeoli.github.io · CV · ORCID · GitLab · LinkedIn
- A. Frangioni, C. Gentile, D. Meoli. Dynamic programming for the single-unit commitment problem with spinning reserves and reactive power. Submitted, 2026.
- D. Fioriti et al. (incl. D. Meoli). Enhancing energy system modelling with advanced mathematical decomposition techniques: feasibility of coupling SMS++ and PyPSA. IEEE EEEIC / I&CPS Europe, 2025. doi:10.1109/EEEIC/ICPSEurope64998.2025.11169291
| SMS++ | C++ framework for block-structured optimization: a model is a tree of Blocks, an algorithm is a Solver that attaches to one. One of its two main maintainers, with Antonio Frangioni. SVMBlock, TwoStageStochasticBlock, MultiStageStochasticBlock, BendersDecompositionSolver and BranchAndXSolver are mine end to end; large parts of UCBlock (including the exact DP solvers for the single-unit commitment problem), of BundleSolver, MILPSolver and the core are as well. |
| aima-python | Fourth-most-active contributor to Russell & Norvig's reference code: CDCL SAT solving, AC3b/AC4/GAC, planning-as-search translators, SVMs, the missing 4th-edition algorithms. |
| optiml | Optimizers, from line search to proximal bundle and interior point, wired to scikit-learn-compatible models trained with them. |
| pypsa2smspp · pySMSpp | Driving SMS++ from Python, and importing models written elsewhere into it. |
Also upstream in SciPy, HiGHS, qpsolvers, MCFClass, FastFlow and the COIN-OR conda-forge feedstocks.