Combinatorial Optimization | Intelligent Optimization Lab

Large Language Models for Optimization

Sat, 26 Sep 2026 00:00:00 +0000

IOLab investigates how large language models can complement established combinatorial optimization techniques. Rather than treating an LLM as a stand-alone solver, we study its role as one component of a structured optimization system.

Our work examines LLM-based generative operators within metaheuristics, the ability of models to recall and interpret problem features, and the information encoded in their internal representations. These representations can be evaluated as surrogates for predicting instance difficulty and algorithm behavior in algorithm-selection scenarios.

The group has published a systematic review of LLMs for combinatorial optimization and experimental work on feature extraction and algorithm selection.

Optimization Methods

Sat, 26 Sep 2026 00:00:00 +0000

IOLab designs and studies algorithms for complex combinatorial optimization problems. Our core expertise includes local-search metaheuristics such as tabu search, simulated annealing, and large neighborhood search, together with multi-neighborhood and hybrid approaches.

We combine these techniques with constraint programming, mixed-integer programming, and automated algorithm configuration. The goal is not only to obtain high-quality solutions, but also to build methods that are robust, reusable, and suitable for the operational constraints of real applications.

A complementary research line studies how optimization algorithms behave. We use statistical experimental design, instance-space analysis, local-optima networks, and algorithm selection to characterize problem instances and understand which methods work best under different conditions.

These methods support the group’s application work in scheduling and timetabling, healthcare, logistics, industrial production, and decision-support systems.

Large Language Models for Combinatorial Optimization: A Systematic Review

Wed, 01 Apr 2026 00:00:00 +0000

Add the full text or supplementary notes for the publication here using Markdown formatting.

Behavior and representation in open-weight Large Language Models for combinatorial optimization: From feature extraction to algorithm selection

Thu, 01 Jan 2026 00:00:00 +0000

Add the full text or supplementary notes for the publication here using Markdown formatting.

ROAR-NET

Sun, 01 Jan 2023 00:00:00 +0000

ROAR-NET — Randomised Optimisation Algorithms Research Network is the European COST Action CA22137, running from 2023 to 2027. The network brings together researchers working on randomized optimization algorithms and promotes shared foundations, tools, and standards across metaheuristics and evolutionary computation.

Luca Di Gaspero served as Leader of Working Group 1 — Problem Modelling and User Experience until 31 March 2026. In this role, he coordinated work on foundational modelling tools and the ROAR-NET API, a shared specification for representing optimization problems and interfacing them with randomized optimization engines. He continues to contribute as a member of the API supervisory committee.

Logistics

Tue, 01 Jun 2021 17:03:29 +0200

IOLab studies complex routing, transportation, and packing problems arising in operational settings. Our research extends classical models such as the Vehicle Routing Problem to account for heterogeneous fleets, periodic services, intermediate facilities, nonlinear charging, uncertain demand, and other constraints found in practice.

Current work focuses on sustainable waste collection, including the planning of electric-vehicle fleets in collaboration with AcegasApsAmga. The activity combines mathematical models, metaheuristics, and data-driven analysis to design efficient collection plans while reducing environmental impact.

The group also works on transportation and service logistics, including railway capacity estimation for industrial freight junctions and location-routing models for emergency medical services.

Open resources

We publish datasets and validation tools for reproducible research, including instances for periodic waste-collection routing and electric-vehicle routing with nonlinear charging. See the IOLab organization on GitHub and the Smart Waste Collection project.

Educational Timetabling

Tue, 01 Jun 2021 17:02:33 +0200

IOLab has worked on automated timetabling since the group was founded in 2000. We develop models and algorithms for assigning lectures, examinations, and other events to times and rooms while balancing hard operational constraints with the preferences and workload of students and staff.

Our research covers local-search metaheuristics, constraint programming, hybrid methods, benchmark design, and rigorous experimental comparison. Group members have also contributed to the organization of the International Timetabling Competitions and to widely used benchmark datasets for course timetabling.

The methods developed by the group have had practical impact beyond research. EasyLocal++, our open-source framework for local-search algorithms, became the technological foundation of EasyCourse, a commercial university timetabling system developed by EasyStaff.

The same expertise extends to sports timetabling: an IOLab team placed second in the Fifth International Timetabling Competition in 2021.

EasyLocal

Wed, 01 Jan 2003 00:00:00 +0000

EasyLocal++ is an open-source C++ framework for the rapid development of local-search algorithms and metaheuristics. It provides reusable abstractions for solution states, neighborhoods, cost components, search runners, and combinations of optimization techniques.

Created at the University of Udine and continuously developed since 2003, the framework has supported research across IOLab’s application domains, from timetabling and scheduling to logistics and healthcare. Its architecture makes it possible to prototype new methods while retaining control over problem-specific data structures and performance.

EasyLocal++ also provided the technological foundation for EasyCourse, the commercial university timetabling system developed by EasyStaff, demonstrating how research software can progress from methodological work to operational use.

The source code and examples are available in the IOLab EasyLocal++ repository. The group also develops JuLES, a Julia framework based on the same approach.