Mathieu Lerouge

About me

Current position

I am currently an Optimization Scientist at DecisionBrain. I focus on modeling operational problems and implementing mathematical optimization solutions to meet clients' specific needs.

Previously, I was a postdoctoral fellow in Computer Science at the Dipartimento di Ingegneria dell'Energia Elettrica e dell'Informazione "Guglielmo Marconi" of the Università di Bologna. My postdoctoral work dealt with Machine-Learning-aided reoptimization methods for real-time adjustments and human interaction.

I hold a PhD in Computer Science from CentraleSupélec - Université Paris-Saclay. My PhD research aimed at developing approaches for modeling and generating explanations about the solutions stemming from optimization systems for their end-users. More specifically, it focused on providing user-centered explanations in the case of the Workforce Scheduling and Routing Problem (WSRP).

Topics of interest

Operations Research, Combinatorial Optimization, Mathematical Programming, Metaheuristics
Artificial Intelligence, Machine Learning
Explanations, Explainable Artificial Intelligence
Ethics and Operations Research, Ethics and Artificial Intelligence, Sustainability
Decision support system

Contact

Email: mathieu [dot] lerouge [at] decisionbrain [dot] com
LinkedIn: mathieulerouge75

Curriculum vitae

Research

Post-doc

Project title: “Machine-Learning-aided reoptimization methods for real-time adjustments.”
Project description: In industrial and operational contexts, optimal solutions to combinatorial optimization problems often become infeasible due to unpredictable disruptions, such as machine breakdowns. Re-solving the underlying Mixed-Integer Linear Programming (MILP) models from scratch is computationally prohibitive for real-time reactions, while simple repair heuristics typically yield poor-quality results. Furthermore, practical adjustments require operational stability, meaning the new plan should not deviate excessively from the original one. This post-doctoral research addresses these challenges by developing a novel "learning-to-reoptimize" framework. Specifically, we design a fix-and-reoptimize strategy aided by a Graph Neural Network (GNN). Given a feature graph encoding the information about the disruption and the solution before disruption, the GNN predicts the likelihood that specific binary decision variables need to be modified. This drastically reduces the MILP search space by identifying a significant subset of decision variables to hard-fix while reoptimizing the rest. Applied to the Lot Sizing Problem (LSP), numerical experiments demonstrate that this graph-based approach successfully generalizes across varying instance sizes and consistently yields good-quality solutions within short time limits.
Institution: Dipartimento di Ingegneria dell'Energia Elettrica e dell'Informazione "Guglielmo Marconi" (DEI), Università di Bologna (UniBo).
In collaboration with: DecisionBrain.
Supervisors: Andrea Lodi (DEI, UniBo), Enrico Malaguti (DEI, UniBo), Michele Monaci (DEI, UniBo), Filippo Focacci (DecisionBrain).
Date: January 2024 - December 2025.

PhD thesis

Manuscript: “Designing and generating user-centered explanations about solutions of a Workforce Scheduling and Routing Problem.”
Abstract: Decision support systems based on combinatorial optimization find application in various professional domains. However, decision-makers who use these systems often lack understanding of their underlying mathematical concepts and algorithmic principles. This knowledge gap can lead to skepticism and reluctance in accepting system-generated solutions, thereby eroding trust in the system. This thesis addresses this issue in the case of the Workforce Scheduling and Routing Problems (WSRP), a combinatorial optimization problem involving human resource allocation and routing decisions. We develop methods for generating explanation texts of different types, with a high-level vocabulary adapted to the end-users. Our methods rely on efficient algorithms for computing and extracting the relevant explanatory information and populates explanation template texts. Numerical experiments show that these algorithms have execution times that are mostly compatible with near-real-time use of explanations by end-users.
Institution: MICS (Mathématiques et Informatique pour la Complexité et les Systèmes), CentraleSupélec - Université Paris-Saclay
Directions: Vincent Mousseau (MICS), Wassila Ouerdane (MICS), Céline Gicquel (LISN).
In collaboration with: DecisionBrain and IBM within the joint public-private (PSPC) project in AI called AIDA.
Date: December 2020 - November 2023.

Publications

“A GNN-aided fix-and-reoptimize approach for the Lot Sizing Problem”, Mathieu Lerouge, Andrea Lodi, Enrico Malaguti, Michele Monaci & Filippo Focacci, Working Paper.
“Modeling and generating user-centered contrastive explanations for the Workforce Scheduling and Routing Problem”, Mathieu Lerouge, Céline Gicquel, Vincent Mousseau & Wassila Ouerdane, ITOR, 2024.
“Counterfactual Explanations for Workforce Scheduling and Routing Problems”, Mathieu Lerouge, Céline Gicquel, Vincent Mousseau & Wassila Ouerdane, Proceedings of the 12th International Conference on Operations Research and Enterprise Systems (ICORES), 2023.

Participation to events

Presentation “A GNN-aided fix-and-reoptimize approach for the Lot Sizing Problem”, at the annual congress in Operations Research and Decision Support ROADEF 2026, organized by the French association ROADEF, held in Tours (France), on February 24th to 26th 2026.
Presentation “ML-guided MILP reoptimization applied to Lot Sizing Problem”, at the annual congress in Operations Research and Decision Support ROADEF 2025, organized by the French association ROADEF, held in Paris (France), on February 26th to 28th 2025.
Presentation “Generating explanations of various types for end-users of optimization systems, application to the Workforce Scheduling and Routing Problem”, at the annual congress in Operations Research and Decision Support ROADEF 2023, organized by the French association ROADEF, held in Rennes (France), on February 20th to 23rd 2023.
Presentation “Generating counterfactual explanations for the Workforce Scheduling and Routing Problem”, at the annual International Conference on Operations Research and Enterprise System (ICORES), held in Lisbon (Portugal), on February 19th to 21st 2023.
Presentation “Designing methods for explaning solutions stemming from optimization systems, application to the Workforce Scheduling and Routing Problem”, at the annual congress in Operations Research and Decision Support ROADEF 2022, organized by the French association ROADEF, held in Lyon (France), on February 23rd to 25th 2022.
Pitch “My PhD subject in less than 6 minutes” [in French], at the annual celebration day of the Doctoral School Interfaces, held at CentraleSupélec in Gif-sur-Yvette (France), on December 6th 2021.
Presentation “Challenges of designing explanation tools for optimization systems”, at the seminar Forum Industriel de l’Intelligence Artificielle (FIIA) organized by the Association Française pour l'Intelligence Artificielle (AFIA) on the topic "Trustworthy AI: responsability, robustness, transparency", held in Paris (France), on October 7th 2021.
Poster presentation “Designing explanation tools for optimization systems - Application to the Workforce Scheduling and Routing Problem”, at the symposium organized by the Groupe De Recherche Intelligence Artificielle (GDRIA) - Institut d’Automne en Intelligence Artificielle (IA2) on the topic “Artificial Intelligence and Explainability”, held in Paris (France), from September 27th to October 1st 2021.

Awards

Best Paper Award Honorable for “Counterfactual Explanations for Workforce Scheduling and Routing Problems” received at the 12th International Conference on Operations Research and Enterprise Systems (ICORES) held in 2023.

Videos

Teaching

Teaching assistant at Università di Bologna

November 2025: Practical works (3 hours) for the Master's degree course “Network optimization”. Course content includes standard algorithms for solving network optimization problems (e.g. Prim's, Dijkstra's, Ford-Fulkerson algorithms). Master degree
October 2024: Lecture (3 hours) for the Master's degree course "Optimization Models and Algorithms". Course content includes MILP decomposition methods (e.g. column generation, Bender's decomposition).

Teaching assistant at CentraleSupélec - Université Paris-Saclay

February 2024 - March 2024: Practical works (30 hours) for the Master's degree course “Decision support: models, algorithms and programming”. Course content includes modeling of decision problems, mathematical programming, multi-objective optimization, metaheuristics and programming using Python & GUROBI.
February 2023 - March 2023: Supervised classes and practical works (60 hours) for the Master's degree course “Decision support: models, algorithms and programming”.
November 2022: Practical works (24 hours) for the Bachelor's degree courses “Coding weeks”. Course content includes Python programming, introduction to git and gitlab, team work.
February 2022 - March 2022: Supervised classes and practical works (60 hours) for the Master's degree course “Decision support: models, algorithms and programming”.
November 2021 - January 2022: Supervised classes (18 hours) for the Bachelor's degree course “Algorithmics and Complexity”. Course content includes data structures, graph algorithms, dynamic programming, computational complexity, complexity theory.
February 2021 - March 2021: Supervised classes and practical works (60 hours) for the Master's degree “Decision support: models, algorithms and programming”.