Michela Milano Home Page

Michela Milano
DISI - Università di Bologna
Viale Risorgimento, 2 - 40136 - Bologna (ITALY)

Tel +39 051 2093790
E-Mail: michela.milano@unibo.it


About me

Michela Milano is full professor at DISI, the Department Computer Science and Engineering of the University of Bologna. She received her Ph.D. in Computer Science in 1998 with a thesis on Constraint Programming. 

Her research interests cover the area of hybrid optimization, a multi-disciplinary field at the cross-road of computer science and applied mathematics, optimization and machine learning and  computational sustainability.

She is Editor in Chief of the Constraint Journal, Associate Editor of ACM Computing Surveys, Area Editor of Constraint Programming Letters and Area Editor of INFORMS Journal on Computing. She has been Associate Editor for INFORMS Journal on Computing and member of the Editorial Board of the Constraint Journal, She has been guest editor for a number of special issues in international journals.

She is elected member of Board of the European Association of Artificial Intelligence (EurAI) , executive councillor of the Associaton for the Advancements of Artificial Intelligence (AAAI), and has been member of the Executive committee of the Association of Constraint Programming (ACP) and the Italian association of Artificial Intelligence (AI*IA).

She is author of 150+ papers on peer reviewed international conferences and journals, editor of four books on hybrid optimizan and guest editor of a three special issues. She is one of the founders of the Constraint Programming and Operations Reseach community, organizer of the first International CPAIOR workshop in Ferrara in 1999 and program chair of CPAIOR 2005 (Prague, Czech Republique) and CPAIOR 2010 (Bologna, Italy). She has been program chair of CP2012 (Quebec City, Canada) and CompSust 2012 (Copenhagen, Denmark).

Michela Milano has a strong track record of fund raising activities collecting more than 2.5M euros in the last five years. She is the recipient of the Google Faculty Research Award on DeepOpt: Embedding deep networks in Combinatorial Optimization. She has been the coordinator of the EU FP7 project e-POLICY - Engineering the POlicy making LIfe CYcle (2011-2014), aimed at integrating optimization and decision support techniques with social simulation and game theory to help policy makers in their decision process. The project also exploits opinion mining and visual analitics techniques. She is partner of the H2020 project OPRECOMP - Open Transprecision Computing (2017-2020), the  EU-FP7 project  COLOMBO - Cooperative Self-Organizing System for low Carbon Mobility at low penetration rates (2012 -2015), of the EU-FP7-Smartcity project DAREED - Decision Support Advidor for innovative business and user engagement for smart energy efficient districts, and of the H2020 FET Proactive - OPRECOMP (2017-2020) Open Transprecision Computing. She has been Italian coordinator for an exchange programme Italy Quebec: Algorithms and systems for the operational planning in industry and services (2007-2009) and partner in a number of Italian projects funded by MIUR and MISE. Michela Milano has a long track record of collaborations with industries and she is among the founders of MindIT.




Editorial and Organization Activities

Editorial Activities

  Conference Chair

 Program committee member



Recent Publications

Export 79 results:
Sort by: Author Title Type [ Year  (Desc)]
Filters: Author is milano  [Clear All Filters]
Galassi, A., M. Lombardi, P. Mello, and M. Milano, "Model Agnostic Solution of CSPs via Deep Learning: A Preliminary Study", Integration of Constraint Programming, Artificial Intelligence, and Operations Research, Cham, Springer International Publishing, pp. 254–262, 06/2018. Abstract
Galassi, A., M. Lombardi, P. Mello, and M. Milano, "Deep Neural Networks for Constraint Satisfaction Problems: an Initial Investigation for the N-Queens", OLA 2018: International Workshop on Optimization and Learning: Challenges and Application, Alicante, Spain, 02/2018.
De Filippo, A., M. Lombardi, M. Milano, and A. Borghetti, "Robust Optimization for Virtual Power Plants", AI*IA 2017 Advances in Artificial Intelligence: XVIth International Conference of the Italian Association for Artificial Intelligence, Bari, Italy, November 14-17, 2017, Proceedings, Cham, Springer International Publishing, pp. 17–30, 2017. Abstract
Bonfietti, A., M. Lombardi, and M. Milano, "Embedding Decision Trees and Random Forests in Constraint Programming", Integration of {AI} and {OR} Techniques in Constraint Programming - 12th International Conference, {CPAIOR} 2015, Barcelona, Spain, May 18-22, 2015, Proceedings, pp. 74–90, 2015.
Borghesi, A., M. Milano, M. Gavanelli, and T. Woods, "Simulation of Incentive mechanisms for renewable energy policies", 27th European Conference on Modelling and Simulation, Alesund, Norway, European Council for Modeling and Simulation , 05/2013.  Download: abs_ECMS2013_0128.pdf (734.65 KB)
Milano, M., "Sustainable Energy Policies: Challenges and Opportunities", Design and Automation Europe - DATE 2013, Grenoble, KP Publications, 03/2013. Abstract
Bonfietti, A., M. Lombardi, and M. Milano, "De-Cycling Cyclic Scheduling Problems", Twenty-Third International Conference on Automated Planning and Scheduling, 2013.
Gavanelli, M., F. Riguzzi, M. Milano, and P. Cagnoli, "Optimization Techniques for Supporting Policy Making", Computational Intelligent Data Analysis for Sustainable Development: Taylor-Francis, pp. 361-380, 2013.  Download: Chapter12.pdf (781.66 KB)
Lombardi, M., M. Milano, and L. Benini, "Robust Scheduling of Task Graphs under Execution Time Uncertainty", IEEE Transactions on Computers, vol. 62, issue 1, no. 99: IEEE, pp. 90--111, 2013.
Kiziltan, Z., A. Lodi, M. Milano, and F. Parisini, "Bounding, Filtering and Diversification in CP-based Local Branching", Journal of Heuristics, vol. 18, issue 3, pp. 353-374, 2012.  Download: JoHLocalBranching.pdf (695.29 KB)
Milano, M., and F. Parisini, "Sliced Neighborhood Search", Expert Systems with Applications, vol. 39, issue 5, pp. 5739-5747 , 2012.
Gavanelli, M., M. Milano, B. O'Sullivan, and A. Holland, "What-if analysis through simulation-optimization hybrids", European Conference on Modeling and Simulation - Track on Policy Modeling, Koblenz, 2012.  Download: ecms12.pdf (1.62 MB)
Gavanelli, M., F. Riguzzi, M. Milano, D. Sottara, A. Cangini, and P. Cagnoli, "An Application of Fuzzy Logic to Strategic Environmental Assessment", XII-th Conference of the Italian Association for Artificial Intelligence, Palermo, Springer Verlag, 2011.

Research Interests


Hybrid Optimization

Hybrid Optimization

Hybrid optimization is a set of multi-disciplinary approaches for solving decision and optimization problems that merge constraint reasoning, applied mathematics, computational logics, metaheuristics and statistics. The motivations for using hybrid optimization is that real problems present a very complex structure, with side  constraints and uncertain data and are composed by distinct, yet tightly connected subproblems. Thus, a single approach is in general ineffective and hybrid solvers are more effective.

Specific topics of interest are

Related links

CPAIOR Conference Series: CPAIOR started as a Workshop in Ferrara in 1999 and became a conference in 2004. Now it is the major forum for researchers interested in combining operations research and artificial intelligence methods in Constraint Programming.

Cost Based Filtering


Constraint Programming global constraints are very effective tools enabling concise modeling. From an operational perspective they can be seen as software components embedding powerful filtering algorithms able to remove provably infeasible portions of the search  space.

In optimization problems, beside feasibility reasoning global constraints can perform optimality reasoning and prune the search space by removing provably sub-optimal parts.

Optimization global constraints embed, beside the traditional filtering algorithm, an optimization component modeling a relaxation of the constraint itself and providing three pieces of information:

  1. the optimal solution of the relaxed problem - x*
  2. the optimal value of the objective function - LB
  3. a set of reduced costs - grad(X,v)

These pieces of information can be used to filter the domain of variables involved in the constraint.



The optimization component can be either a linear relaxation possibly tighted with cutting planes or a combinatorial relaxation.


Logic Based Benders Decomposition

The classical Benders Decomposition method decomposes a problem into two loosely connected subproblems. It enumerates values for the connecting variables. For each set of values enumerated, it solves the subproblem that results from fixing the connecting variables to these values. The solution of the subproblem generates a Benders cut that the connecting variables must satisfy in all subsequent solutions enumerated.

The process continues until the master problem and subproblem converge providing the same value. The classical Benders approach, however, requires that the subproblem is a continuous linear or nonlinear programming problem.

Logic-Based Benders Decomposition LBBD is an extension of the tranditional scheme that enables generic solvers to be used as subproblem solvers.

We have applied LBDD to scheduling problems with alternative resources in the field of embedded system design. The structure of a solver exploits problem decomposition, where task to resource assignments influence the objective function and are performed via an Integer Linear Programming  (ILP) solver. The scheduling part (with fixed resources), instead, is better dealt with Constraint Programming (CP).


The figure depicts the overall process. Resource allocation is computed by optimizing the objective function. The valid allocation is passed to the scheduling component. If a schedule exist for such allocation, the process converges to the optimal solution. Otherwise a no good is generated (as a linear constraint) and passed to the resource allocation component.

Two aspects influence the overall process: (1) to avoid the generation of trivially infeasible solutions in the allocation part, the ILP model should embed a scheduling problem relaxation; (2) tight no-goods can heavily reduce the number of iterations of the overall solution process.

We have faced different problems with this structure: those where the objective function depends only on the resource allocation variables, and those where also scheduling decision impact the objective function. We have deeply investigated tight relaxation and tight cuts to speed up the solution process.


Research projects





Teaching Activities

 Corsi 2017/2018

Corsi 2016/2017

Corsi 2015/2016

Corsi 2013/2014

Corsi 2012/2013

Corsi 2011/2012

Corsi 2010/2011

Corsi 2009/2010


Corso su Global System Science e Smart Cities

Corsi Vecchio Ordinamento

La prof. Milano svolge gli esami di Fondamenti di Informatica LA vecchio ordinamento per Ing. Elettronica, Ing. delle Telecomunicazioni e Ing. Automatica. Gli studenti possono trovare il programma e il materiale per sostenere l'esame al seguente link