Distributed Estimation of Distribution Algorithms for continuous optimization: how does the exchanged information influence their behavior?

One of the most promising areas in which probabilistic graphical models have shown an incipient activity is the field of heuristic optimization and, in particular, in Estimation of Distribution Algorithms. Due to their inherent parallelism, different research lines have been studied trying to improve Estimation of Distribution Algorithms from the point of view of execution time and/or accuracy. Among these proposals, we focus on the so-called distributed or island-based models. This approach defines several islands (algorithms instances) running independently and exchanging information with a given frequency. The information sent by the islands can be either a set of individuals or a probabilistic model. This paper presents a comparative study for a distributed univariate Estimation of Distribution Algorithm and a multivariate version, paying special attention to the comparison of two alternative methods for exchanging information, over a wide set of parameters and problems ? the standard benchmark developed for the IEEE Workshop on Evolutionary Algorithms and other Metaheuristics for Continuous Optimization Problems of the ISDA 2009 Conference. Several analyses from different points of view have been conducted to analyze both the influence of the parameters and the relationships between them including a characterization of the configurations according to their behavior on the proposed benchmark.

The optimization principle in phylogenetic analysis tends to give incorrect topologies when the number of nucleotides or amino acids used is small

In the maximum parsimony (MP) and minimum evolution (ME) methods of phylogenetic inference, evolutionary trees are constructed by searching for the topology that shows the minimum number of mutational changes required (M) and the smallest sum of branch lengths (S), respectively, whereas in the maximum likelihood (ML) method the topology showing the highest maximum likelihood (A) of observing a given data set is chosen. However, the theoretical basis of the optimization principle remains unclear. We therefore examined the relationships of M, S, and A for the MP, ME, and ML trees with those for the true tree by using computer simulation. The results show that M and S are generally greater for the true tree than for the MP and ME trees when the number of nucleotides examined (n) is relatively small, whereas A is generally lower for the true tree than for the ML tree. This finding indicates that the optimization principle tends to give incorrect topologies when n is small. To deal with this disturbing property of the optimization principle, we suggest that more attention should be given to testing the statistical reliability of an estimated tree rather than to finding the optimal tree with excessive efforts. When a reliability test is conducted, simplified MP, ME, and ML algorithms such as the neighbor-joining method generally give conclusions about phylogenetic inference very similar to those obtained by the more extensive tree search algorithms.

Assessing sequence comparison methods with reliable structurally identified distant evolutionary relationships

Pairwise sequence comparison methods have been assessed using proteins whose relationships are known reliably from their structures and functions, as described in the scop database [Murzin, A. G., Brenner, S. E., Hubbard, T. & Chothia C. (1995) J. Mol. Biol. 247, 536–540]. The evaluation tested the programs blast [Altschul, S. F., Gish, W., Miller, W., Myers, E. W. & Lipman, D. J. (1990). J. Mol. Biol. 215, 403–410], wu-blast2 [Altschul, S. F. & Gish, W. (1996) Methods Enzymol. 266, 460–480], fasta [Pearson, W. R. & Lipman, D. J. (1988) Proc. Natl. Acad. Sci. USA 85, 2444–2448], and ssearch [Smith, T. F. & Waterman, M. S. (1981) J. Mol. Biol. 147, 195–197] and their scoring schemes. The error rate of all algorithms is greatly reduced by using statistical scores to evaluate matches rather than percentage identity or raw scores. The E-value statistical scores of ssearch and fasta are reliable: the number of false positives found in our tests agrees well with the scores reported. However, the P-values reported by blast and wu-blast2 exaggerate significance by orders of magnitude. ssearch, fasta ktup = 1, and wu-blast2 perform best, and they are capable of detecting almost all relationships between proteins whose sequence identities are >30%. For more distantly related proteins, they do much less well; only one-half of the relationships between proteins with 20–30% identity are found. Because many homologs have low sequence similarity, most distant relationships cannot be detected by any pairwise comparison method; however, those which are identified may be used with confidence.

Population-based continuous optimization, probabilistic modelling and mean shift

Evolutionary algorithms perform optimization using a population of sample solution points. An interesting development has been to view population-based optimization as the process of evolving an explicit, probabilistic model of the search space. This paper investigates a formal basis for continuous, population-based optimization in terms of a stochastic gradient descent on the Kullback-Leibler divergence between the model probability density and the objective function, represented as an unknown density of assumed form. This leads to an update rule that is related and compared with previous theoretical work, a continuous version of the population-based incremental learning algorithm, and the generalized mean shift clustering framework. Experimental results are presented that demonstrate the dynamics of the new algorithm on a set of simple test problems.

What are dynamic optimization problems?

Dynamic Optimization Problems (DOPs) have been widely studied using Evolutionary Algorithms (EAs). Yet, a clear and rigorous definition of DOPs is lacking in the Evolutionary Dynamic Optimization (EDO) community. In this paper, we propose a unified definition of DOPs based on the idea of multiple-decision-making discussed in the Reinforcement Learning (RL) community. We draw a connection between EDO and RL by arguing that both of them are studying DOPs according to our definition of DOPs. We point out that existing EDO or RL research has been mainly focused on some types of DOPs. A conceptualized benchmark problem, which is aimed at the systematic study of various DOPs, is then developed. Some interesting experimental studies on the benchmark reveal that EDO and RL methods are specialized in certain types of DOPs and more importantly new algorithms for DOPs can be developed by combining the strength of both EDO and RL methods.

Temperature management for heterogeneous multi-core FPGAs using adaptive evolutionary multi-objective approaches

Heterogeneous multi-core FPGAs contain different types of cores, which can improve efficiency when used with an effective online task scheduler. However, it is not easy to find the right cores for tasks when there are multiple objectives or dozens of cores. Inappropriate scheduling may cause hot spots which decrease the reliability of the chip. Given that, our research builds a simulating platform to evaluate all kinds of scheduling algorithms on a variety of architectures. On this platform, we provide an online scheduler which uses multi-objective evolutionary algorithm (EA). Comparing the EA and current algorithms such as Predictive Dynamic Thermal Management (PDTM) and Adaptive Temperature Threshold Dynamic Thermal Management (ATDTM), we find some drawbacks in previous work. First, current algorithms are overly dependent on manually set constant parameters. Second, those algorithms neglect optimization for heterogeneous architectures. Third, they use single-objective methods, or use linear weighting method to convert a multi-objective optimization into a single-objective optimization. Unlike other algorithms, the EA is adaptive and does not require resetting parameters when workloads switch from one to another. EAs also improve performance when used on heterogeneous architecture. A efficient Pareto front can be obtained with EAs for the purpose of multiple objectives.

The evolutionary relationships and age of Homo naledi : an assessment using dated Bayesian phylogenetic methods

Acknowledgements We wish to express our gratitude to the National Geographic Society and the National Research Foundation of South Africa for funding the discovery, recovery, and analysis of the H. naledi material. The study reported here was also made possible by grants from the Social Sciences and Humanities Research Council of Canada, the Canada Foundation for Innovation, the British Columbia Knowledge Development Fund, the Canada Research Chairs Program, Simon Fraser University, the DST/NRF Centre of Excellence in Palaeosciences (COE-Pal), as well as by a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada, a Young Scientist Development Grant from the Paleontological Scientific Trust (PAST), a Baldwin Fellowship from the L.S.B. Leakey Foundation, and a Seed Grant and a Cornerstone Faculty Fellowship from the Texas A&M University College of Liberal Arts. We would like to thank the South African Heritage Resource Agency for the permits necessary to work on the Rising Star site; the Jacobs family for granting access; Wilma Lawrence, Bonita De Klerk, Merrill Van der Walt, and Justin Mukanku for their assistance during all phases of the project; Lucas Delezene for valuable discussion on the dental characters of H. naledi. We would also like to thank Peter Schmid for the preparation of the Dinaledi fossil material; Yoel Rak for explaining in detail some of the characters used in previous studies; William Kimbel for drawing our attention to the possibility that there might be a problem with Dembo et al.’s (2015) codes for the two characters related to the articular eminence; Will Stein for helpful discussion about the Bayesian analyses; Mike Lee for his comments on this manuscript; John Hawks for his support in organizing the Rising Star workshop; and the associate editor and three anonymous reviewers for their valuable comments. We are grateful to S. Potze and the Ditsong Museum, B. Billings and the School of Anatomical Sciences at the University of the Witwatersrand, and B. Zipfel and the Evolutionary Studies Institute at the University of the Witwatersrand for providing access to the specimens in their care; the University of the Witwatersrand, the Evolutionary Studies Institute, and the South African National Centre of Excellence in PalaeoSciences for hosting a number of the authors while studying the material; and the Western Canada Research Grid for providing access to the high-performance computing facilities for the Bayesian analyses. Last but definitely not least, we thank the head of the Rising Star project, Lee Berger, for his leadership and support, and for encouraging us to pursue the study reported here.

Aerodynamic optimization using Adjoint methods and parametric CAD models

This paper describes an implementation of a method capable of integrating parametric, feature based, CAD models based on commercial software (CATIA) with the SU2 software framework. To exploit the adjoint based methods for aerodynamic optimisation within the SU2, a formulation to obtain geometric sensitivities directly from the commercial CAD parameterisation is introduced, enabling the calculation of gradients with respect to CAD based design variables. To assess the accuracy and efficiency of the alternative approach, two aerodynamic optimisation problems are investigated: an inviscid, 3D, problem with multiple constraints, and a 2D high-lift aerofoil, viscous problem without any constraints. Initial results show the new parameterisation obtaining reliable optimums, with similar levels of performance of the software native parameterisations. In the final paper, details of computing CAD sensitivities will be provided, including accuracy as well as linking geometric sensitivities to aerodynamic objective functions and constraints; the impact in the robustness of the overall method will be assessed and alternative parameterisations will be included.

Optimization models and methods for real-time transportation planning in forestry

Lors du transport du bois de la forêt vers les usines, de nombreux événements imprévus peuvent se produire, événements qui perturbent les trajets prévus (par exemple, en raison des conditions météo, des feux de forêt, de la présence de nouveaux chargements, etc.). Lorsque de tels événements ne sont connus que durant un trajet, le camion qui accomplit ce trajet doit être détourné vers un chemin alternatif. En l’absence d’informations sur un tel chemin, le chauffeur du camion est susceptible de choisir un chemin alternatif inutilement long ou pire, qui est lui-même "fermé" suite à un événement imprévu. Il est donc essentiel de fournir aux chauffeurs des informations en temps réel, en particulier des suggestions de chemins alternatifs lorsqu’une route prévue s’avère impraticable. Les possibilités de recours en cas d’imprévus dépendent des caractéristiques de la chaîne logistique étudiée comme la présence de camions auto-chargeurs et la politique de gestion du transport. Nous présentons trois articles traitant de contextes d’application différents ainsi que des modèles et des méthodes de résolution adaptés à chacun des contextes. Dans le premier article, les chauffeurs de camion disposent de l’ensemble du plan hebdomadaire de la semaine en cours. Dans ce contexte, tous les efforts doivent être faits pour minimiser les changements apportés au plan initial. Bien que la flotte de camions soit homogène, il y a un ordre de priorité des chauffeurs. Les plus prioritaires obtiennent les volumes de travail les plus importants. Minimiser les changements dans leurs plans est également une priorité. Étant donné que les conséquences des événements imprévus sur le plan de transport sont essentiellement des annulations et/ou des retards de certains voyages, l’approche proposée traite d’abord l’annulation et le retard d’un seul voyage, puis elle est généralisée pour traiter des événements plus complexes. Dans cette ap- proche, nous essayons de re-planifier les voyages impactés durant la même semaine de telle sorte qu’une chargeuse soit libre au moment de l’arrivée du camion à la fois au site forestier et à l’usine. De cette façon, les voyages des autres camions ne seront pas mo- difiés. Cette approche fournit aux répartiteurs des plans alternatifs en quelques secondes. De meilleures solutions pourraient être obtenues si le répartiteur était autorisé à apporter plus de modifications au plan initial. Dans le second article, nous considérons un contexte où un seul voyage à la fois est communiqué aux chauffeurs. Le répartiteur attend jusqu’à ce que le chauffeur termine son voyage avant de lui révéler le prochain voyage. Ce contexte est plus souple et offre plus de possibilités de recours en cas d’imprévus. En plus, le problème hebdomadaire peut être divisé en des problèmes quotidiens, puisque la demande est quotidienne et les usines sont ouvertes pendant des périodes limitées durant la journée. Nous utilisons un modèle de programmation mathématique basé sur un réseau espace-temps pour réagir aux perturbations. Bien que ces dernières puissent avoir des effets différents sur le plan de transport initial, une caractéristique clé du modèle proposé est qu’il reste valable pour traiter tous les imprévus, quelle que soit leur nature. En effet, l’impact de ces événements est capturé dans le réseau espace-temps et dans les paramètres d’entrée plutôt que dans le modèle lui-même. Le modèle est résolu pour la journée en cours chaque fois qu’un événement imprévu est révélé. Dans le dernier article, la flotte de camions est hétérogène, comprenant des camions avec des chargeuses à bord. La configuration des routes de ces camions est différente de celle des camions réguliers, car ils ne doivent pas être synchronisés avec les chargeuses. Nous utilisons un modèle mathématique où les colonnes peuvent être facilement et naturellement interprétées comme des itinéraires de camions. Nous résolvons ce modèle en utilisant la génération de colonnes. Dans un premier temps, nous relaxons l’intégralité des variables de décision et nous considérons seulement un sous-ensemble des itinéraires réalisables. Les itinéraires avec un potentiel d’amélioration de la solution courante sont ajoutés au modèle de manière itérative. Un réseau espace-temps est utilisé à la fois pour représenter les impacts des événements imprévus et pour générer ces itinéraires. La solution obtenue est généralement fractionnaire et un algorithme de branch-and-price est utilisé pour trouver des solutions entières. Plusieurs scénarios de perturbation ont été développés pour tester l’approche proposée sur des études de cas provenant de l’industrie forestière canadienne et les résultats numériques sont présentés pour les trois contextes.

Optimization models and methods for real-time transportation planning in forestry

Lors du transport du bois de la forêt vers les usines, de nombreux événements imprévus peuvent se produire, événements qui perturbent les trajets prévus (par exemple, en raison des conditions météo, des feux de forêt, de la présence de nouveaux chargements, etc.). Lorsque de tels événements ne sont connus que durant un trajet, le camion qui accomplit ce trajet doit être détourné vers un chemin alternatif. En l’absence d’informations sur un tel chemin, le chauffeur du camion est susceptible de choisir un chemin alternatif inutilement long ou pire, qui est lui-même "fermé" suite à un événement imprévu. Il est donc essentiel de fournir aux chauffeurs des informations en temps réel, en particulier des suggestions de chemins alternatifs lorsqu’une route prévue s’avère impraticable. Les possibilités de recours en cas d’imprévus dépendent des caractéristiques de la chaîne logistique étudiée comme la présence de camions auto-chargeurs et la politique de gestion du transport. Nous présentons trois articles traitant de contextes d’application différents ainsi que des modèles et des méthodes de résolution adaptés à chacun des contextes. Dans le premier article, les chauffeurs de camion disposent de l’ensemble du plan hebdomadaire de la semaine en cours. Dans ce contexte, tous les efforts doivent être faits pour minimiser les changements apportés au plan initial. Bien que la flotte de camions soit homogène, il y a un ordre de priorité des chauffeurs. Les plus prioritaires obtiennent les volumes de travail les plus importants. Minimiser les changements dans leurs plans est également une priorité. Étant donné que les conséquences des événements imprévus sur le plan de transport sont essentiellement des annulations et/ou des retards de certains voyages, l’approche proposée traite d’abord l’annulation et le retard d’un seul voyage, puis elle est généralisée pour traiter des événements plus complexes. Dans cette ap- proche, nous essayons de re-planifier les voyages impactés durant la même semaine de telle sorte qu’une chargeuse soit libre au moment de l’arrivée du camion à la fois au site forestier et à l’usine. De cette façon, les voyages des autres camions ne seront pas mo- difiés. Cette approche fournit aux répartiteurs des plans alternatifs en quelques secondes. De meilleures solutions pourraient être obtenues si le répartiteur était autorisé à apporter plus de modifications au plan initial. Dans le second article, nous considérons un contexte où un seul voyage à la fois est communiqué aux chauffeurs. Le répartiteur attend jusqu’à ce que le chauffeur termine son voyage avant de lui révéler le prochain voyage. Ce contexte est plus souple et offre plus de possibilités de recours en cas d’imprévus. En plus, le problème hebdomadaire peut être divisé en des problèmes quotidiens, puisque la demande est quotidienne et les usines sont ouvertes pendant des périodes limitées durant la journée. Nous utilisons un modèle de programmation mathématique basé sur un réseau espace-temps pour réagir aux perturbations. Bien que ces dernières puissent avoir des effets différents sur le plan de transport initial, une caractéristique clé du modèle proposé est qu’il reste valable pour traiter tous les imprévus, quelle que soit leur nature. En effet, l’impact de ces événements est capturé dans le réseau espace-temps et dans les paramètres d’entrée plutôt que dans le modèle lui-même. Le modèle est résolu pour la journée en cours chaque fois qu’un événement imprévu est révélé. Dans le dernier article, la flotte de camions est hétérogène, comprenant des camions avec des chargeuses à bord. La configuration des routes de ces camions est différente de celle des camions réguliers, car ils ne doivent pas être synchronisés avec les chargeuses. Nous utilisons un modèle mathématique où les colonnes peuvent être facilement et naturellement interprétées comme des itinéraires de camions. Nous résolvons ce modèle en utilisant la génération de colonnes. Dans un premier temps, nous relaxons l’intégralité des variables de décision et nous considérons seulement un sous-ensemble des itinéraires réalisables. Les itinéraires avec un potentiel d’amélioration de la solution courante sont ajoutés au modèle de manière itérative. Un réseau espace-temps est utilisé à la fois pour représenter les impacts des événements imprévus et pour générer ces itinéraires. La solution obtenue est généralement fractionnaire et un algorithme de branch-and-price est utilisé pour trouver des solutions entières. Plusieurs scénarios de perturbation ont été développés pour tester l’approche proposée sur des études de cas provenant de l’industrie forestière canadienne et les résultats numériques sont présentés pour les trois contextes.

Efficient Hill Climber for Constrained Pseudo-Boolean Optimization Problems

Efficient hill climbers have been recently proposed for single- and multi-objective pseudo-Boolean optimization problems. For $k$-bounded pseudo-Boolean functions where each variable appears in at most a constant number of subfunctions, it has been theoretically proven that the neighborhood of a solution can be explored in constant time. These hill climbers, combined with a high-level exploration strategy, have shown to improve state of the art methods in experimental studies and open the door to the so-called Gray Box Optimization, where part, but not all, of the details of the objective functions are used to better explore the search space. One important limitation of all the previous proposals is that they can only be applied to unconstrained pseudo-Boolean optimization problems. In this work, we address the constrained case for multi-objective $k$-bounded pseudo-Boolean optimization problems. We find that adding constraints to the pseudo-Boolean problem has a linear computational cost in the hill climber.

Multi-objective optimization of material model parameters of an adhesive layer by using SPEA2

The usage of multi material structures in industry, especially in the automotive industry are increasing. To overcome the difficulties in joining these structures, adhesives have several benefits over traditional joining methods. Therefore, accurate simulations of the entire process of fracture including the adhesive layer is crucial. In this paper, material parameters of a previously developed meso mechanical finite element (FE) model of a thin adhesive layer are optimized using the Strength Pareto Evolutionary Algorithm (SPEA2). Objective functions are defined as the error between experimental data and simulation data. The experimental data is provided by previously performed experiments where an adhesive layer was loaded in monotonically increasing peel and shear. Two objective functions are dependent on 9 model parameters (decision variables) in total and are evaluated by running two FEsimulations, one is loading the adhesive layer in peel and the other in shear. The original study converted the two objective functions into one function that resulted in one optimal solution. In this study, however, a Pareto frontis obtained by employing the SPEA2 algorithm. Thus, more insight into the material model, objective functions, optimal solutions and decision space is acquired using the Pareto front. We compare the results and show good agreement with the experimental data.

Evolutionary algorithms for the modeling of bioreactors

This work aims to study the application of Genetic Algorithms in anaerobic digestion modeling, in particular when using dynamical models. Along the work, different types of bioreactors are shown, such as batch, semi-batch and continuous, as well as their mathematical modeling. The work intendeds to estimate the parameter values of two biological reaction model. For that, simulated results, where only one output variable, the produced biogas, is known, are fitted to the model results. For this reason, the problems associated with reverse optimization are studied, using some graphics that provide clues to the sensitivity and identifiability associated with the problem. Particular solutions obtained by the identifiability analysis using GENSSI and DAISY softwares are also presented. Finally, the optimization is performed using genetic algorithms. During this optimization the need to improve the convergence of genetic algorithms was felt. This need has led to the development of an adaptation of the genetic algorithms, which we called Neighbored Genetic Algorithms (NGA1 and NGA2). In order to understand if this new approach overcomes the Basic Genetic Algorithms (BGA) and achieves the proposed goals, a study of 100 full optimization runs for each situation was further developed. Results show that NGA1 and NGA2 are statistically better than BGA. However, because it was not possible to obtain consistent results, the Nealder-Mead method was used, where the initial guesses were the estimated results from GA; Algoritmos Evolucionários para a Modelação de Bioreactores Resumo: Neste trabalho procura-se estudar os algoritmos genéticos com aplicação na modelação da digestão anaeróbia e, em particular, quando se utilizam modelos dinâmicos. Ao longo do mesmo, são apresentados diferentes tipos de bioreactores, como os batch, semi-batch e contínuos, bem como a modelação matemática dos mesmos. Neste trabalho procurou-se estimar o valor dos parâmetros que constam num modelo de digestão anaeróbia para o ajustar a uma situação simulada onde apenas se conhece uma variável de output, o biogas produzido. São ainda estudados os problemas associados à optimização inversa com recurso a alguns gráficos que fornecem pistas sobre a sensibilidade e identifiacabilidade associadas ao problema da modelação da digestão anaeróbia. São ainda apresentadas soluções particulares de idenficabilidade obtidas através dos softwares GENSSI e DAISY. Finalmente é realizada a optimização do modelo com recurso aos algoritmos genéticos. No decorrer dessa optimização sentiu-se a necessidade de melhorar a convergência e, portanto, desenvolveu-se ainda uma adaptação dos algoritmos genéticos a que se deu o nome de Neighboured Genetic Algorithms (NGA1 e NGA2). No sentido de se compreender se as adaptações permitiam superar os algoritmos genéticos básicos e atingir as metas propostas, foi ainda desenvolvido um estudo em que o processo de optimização foi realizado 100 vezes para cada um dos métodos, o que permitiu concluir, estatisticamente, que os BGA foram superados pelos NGA1 e NGA2. Ainda assim, porque não foi possivel obter consistência nos resultados, foi usado o método de Nealder-Mead utilizado como estimativa inicial os resultados obtidos pelos algoritmos genéticos.

Advanced decision-support methods for the design, control, and optimization of perishable products life cycle

In the last decades, global food supply chains had to deal with the increasing awareness of the stakeholders and consumers about safety, quality, and sustainability. In order to address these new challenges for food supply chain systems, an integrated approach to design, control, and optimize product life cycle is required. Therefore, it is essential to introduce new models, methods, and decision-support platforms tailored to perishable products. This thesis aims to provide novel practice-ready decision-support models and methods to optimize the logistics of food items with an integrated and interdisciplinary approach. It proposes a comprehensive review of the main peculiarities of perishable products and the environmental stresses accelerating their quality decay. Then, it focuses on top-down strategies to optimize the supply chain system from the strategical to the operational decision level. Based on the criticality of the environmental conditions, the dissertation evaluates the main long-term logistics investment strategies to preserve products quality. Several models and methods are proposed to optimize the logistics decisions to enhance the sustainability of the supply chain system while guaranteeing adequate food preservation. The models and methods proposed in this dissertation promote a climate-driven approach integrating climate conditions and their consequences on the quality decay of products in innovative models supporting the logistics decisions. Given the uncertain nature of the environmental stresses affecting the product life cycle, an original stochastic model and solving method are proposed to support practitioners in controlling and optimizing the supply chain systems when facing uncertain scenarios. The application of the proposed decision-support methods to real case studies proved their effectiveness in increasing the sustainability of the perishable product life cycle. The dissertation also presents an industry application of a global food supply chain system, further demonstrating how the proposed models and tools can be integrated to provide significant savings and sustainability improvements.

Methods for integrating machine learning and constrained optimization

In the framework of industrial problems, the application of Constrained Optimization is known to have overall very good modeling capability and performance and stands as one of the most powerful, explored, and exploited tool to address prescriptive tasks. The number of applications is huge, ranging from logistics to transportation, packing, production, telecommunication, scheduling, and much more. The main reason behind this success is to be found in the remarkable effort put in the last decades by the OR community to develop realistic models and devise exact or approximate methods to solve the largest variety of constrained or combinatorial optimization problems, together with the spread of computational power and easily accessible OR software and resources. On the other hand, the technological advancements lead to a data wealth never seen before and increasingly push towards methods able to extract useful knowledge from them; among the data-driven methods, Machine Learning techniques appear to be one of the most promising, thanks to its successes in domains like Image Recognition, Natural Language Processes and playing games, but also the amount of research involved. The purpose of the present research is to study how Machine Learning and Constrained Optimization can be used together to achieve systems able to leverage the strengths of both methods: this would open the way to exploiting decades of research on resolution techniques for COPs and constructing models able to adapt and learn from available data. In the first part of this work, we survey the existing techniques and classify them according to the type, method, or scope of the integration; subsequently, we introduce a novel and general algorithm devised to inject knowledge into learning models through constraints, Moving Target. In the last part of the thesis, two applications stemming from real-world projects and done in collaboration with Optit will be presented.