tag:www.gerad.ca,2005:/fr/eventsCalendrier des activités du GERAD2023-12-12T10:50:30-05:0020tag:www.gerad.ca,2005:Event/21262024-07-21T00:00:00-04:002024-07-21T00:00:00-04:0025th International Symposium on Mathematical Programming (ISMP 2024)<p>Conférence</p>
<p><a href="https://ismp2024.gerad.ca/">https://ismp2024.gerad.ca/</a></p>
2023-12-12 10:50:30 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/21912024-07-19T00:00:00-04:002024-07-19T00:00:00-04:00JuMP-dev 2024<p><a href="https://jump.dev/meetings/jumpdev2024/">https://jump.dev/meetings/jumpdev2024/</a></p>
2023-11-16 11:56:46 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/21952024-07-09T00:00:00-04:002024-07-09T00:00:00-04:0020th International Symposium on Dynamic Games and Applications<p>Conférence</p>
<p><a href="https://www.gerad.ca/colloques/isdg2024/">https://www.gerad.ca/colloques/isdg2024/</a></p>
2024-01-17 15:35:52 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/21942024-05-13T00:00:00-04:002024-05-13T00:00:00-04:00Quatorzième atelier de résolution de problèmes industriels de Montréal<p>Atelier</p>
<p>Centre de recherches mathématiques (CRM)<br>
GERAD<br>
IVADO - Consortium de recherche, de formation et de mobilisation des connaissances en intelligence artificielle</p>
<p><a href="https://www.crmath.ca/activites/#/type/activity/id/3955">https://www.crmath.ca/activites/#/type/activity/id/3955</a></p>
2024-01-22 10:08:07 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/21372024-05-06T00:00:00-04:002024-05-06T00:00:00-04:00Journées de l'optimisation 2024<p>Conférence</p>
<p><a href="https://symposia.gerad.ca/jopt2024/fr/">https://symposia.gerad.ca/jopt2024/fr/</a></p>
2023-11-20 13:46:06 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22242024-05-03T00:00:00-04:002024-05-03T00:00:00-04:00Jeux et optimisation - Colloque en l'honneur de Michèle Breton<p>Atelier</p>
<p>Chaire de théorie des jeux et gestion</p>
<p><a href="https://www.gerad.ca/colloques/Michele-Breton-workshop-2024/index-fr.html">https://www.gerad.ca/colloques/Michele-Breton-workshop-2024/index-fr.html</a></p>
2024-02-26 14:03:50 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22282024-04-25T11:00:00-04:002024-04-25T11:00:00-04:00Statistical discrimination without knowing statistics: blame social interactions?<p>Séminaire sur les jeux dynamiques et les applications</p>
<p>Emily Tanimura, Université Paris 1, France</p>
<p>Chaire de théorie des jeux et gestion</p>
<p><img src="/system/assets/000/002/069/2069.TANIMURA-Emily_card.png" alt="Emily Tanimura"></p>
<p><strong>Séminaire en format hybride au GERAD local 4488 ou <a href="https://hecmontreal.zoom.us/j/82911900343?pwd=ZnpodHo4UFVQQ05Tbi96Q0wrMzFmUT09">Zoom</a>.</strong></p>
<p>We consider a model where decision makers repeatedly receive candidates and assign to them a binary decision that we can interpret as hire/not hire. The decision makers base their decision on the characteristics of the candidate but they are also sensitive to the social influence exerted by the observed past choices of their peers. We characterize the long run frequency of decisions in the model, and show in particular that for candidates belonging to a group with ”un- favorable” characteristics, the dynamics increase the rejection rate compared to a scenario with independent decisions, suggesting that social influence between decision makers can generate effects very similar to those that result from statistical discrimination. We then analyze how the existence and magnitude of a reinforcement in rejection rates depend on different properties of the distribution of characteristics in the candidate population. </p>
2024-03-26 14:55:20 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22012024-04-24T11:00:00-04:002024-04-24T11:00:00-04:00Learning and Stochastic Optimization for Industrial Mining Complexes<p>Séminaire “Un(e) chercheur(-euse) du GERAD vous parle!”</p>
<p>Yassine Yaakoubi, Université McGill, Canada</p>
<p>Chaire de recherche du Canada sur le développement durable des ressources minérales et l'optimisation en cas d'incertitude</p>
<p><img src="/system/assets/000/001/386/1386.YaakoubiYassine_card.png" alt="Yassine Yaakoubi"></p>
<p><strong>Séminaire en format hybride au GERAD local 4488 ou <a href="https://hecmontreal.zoom.us/j/81777698995?pwd=WXplZGdqenhLbGJkSnBKSGtBaFpjdz09">Zoom</a>.</strong></p>
<p>Decision-making in complex and/or stochastic settings, such as industrial mining complexes, presents significant challenges across various sectors. This seminar will delve into a novel solution approach that integrates machine learning and optimization for mineral supply/value chains under supply and demand uncertainties. We will explore three interconnected paradigm shifts: 1) data-driven hyper-heuristics with a heuristic search tree and self-adaptive framework; 2) smart lifelong learning context-aware solvers, with a hyper-heuristic for dynamic mining complex modeling, neural diving policy for heuristic selection, and neural branching policy with a soft branching strategy; and 3) a distributional perspective for warm-starting, using historical solutions and graphical models for initial production schedules. The performance of the proposed approach will be demonstrated through computational results and case studies, showcasing up to three orders of magnitude reductions in primal suboptimality and execution times, and increased robustness in solutions yielding up to 40% higher net present values. Finally, we will discuss insights, potential limitations, and future research directions in developing robust, reasoning, and responsible decision-support systems for industrial-scale uncertain environments.</p>
2024-01-03 10:14:58 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22202024-04-22T08:30:00-04:002024-04-22T08:30:00-04:00IVADO PRF3 Workshop 2024: Towards Resilient and Sustainable Supply Chains Through Machine Learning and Optimization<p>Conférence</p>
<p>IVADO - Consortium de recherche, de formation et de mobilisation des connaissances en intelligence artificielle<br>
Chaire de recherche du Canada en analytique de la chaine d’approvisionnement<br>
Chaire de recherche du Canada sur la prise de décision en incertitude<br>
Chaire de recherche du Canada en prévision de la demande et optimisation des systèmes de transport<br>
GERAD<br>
CIRRELT</p>
<p><strong>Vous avez jusqu'au 7 avril pour vous inscrire ici pour participer à la conférence. Frais symboliques : 25 $.
<a href="https://symposia.gerad.ca/April22-conference/register">Veuillez vous inscrire ici.</a></strong></p>
<p>Recent developments in machine learning and optimization have the potential to enable supply chain decision-makers and practitioners to leverage data and deal with uncertainty in an effective and adaptive fashion. This scientific workshop presents various industrial realizations demonstrating how modern machine learning and optimization methods can effectively be applied in practice. Through academic-style talks from speakers in both academia and industry, they will present various use cases in supply chains and scientific approaches, as well as challenges and opportunities in the applications of learning and optimization methods. There will also be dedicated time for informal discussions among participants.</p>
<p>The conference provides coffee, lunch and post-event cocktail reception with finger foods.</p>
<hr>
<h4>Titles and speakers:</h4>
<ul>
<li><p><strong>Practical Considerations for Decision Under Uncertainty in an Industrial Setting</strong><br>
Marie-Claude Côté, Sacha Izadi, Nicolas Boez, Jean-François Landry (IVADO Labs)</p></li>
<li><p><strong>Sample Efficient Algorithms for Urban Mobility Problems Through Physics-Informed Machine Learning</strong><br>
Carolina Osorio (HEC Montréal and Google Research)</p></li>
<li><p><strong>Ship Route Optimization - Multi-Objective Approach</strong><br>
Capt. Gurjeet Warya (True North Marine)</p></li>
<li><p><strong>Humanitarian Supply Chain Analytics</strong><br>
Marie-Ève Rancourt (HEC Montréal)</p></li>
<li><p><strong>Operational Design and Data in Smallholder Farmer Supply Chains</strong><br>
Joann de Zegher (PemPem)</p></li>
<li><p><strong>Optimize-Then-Predict: An Imitation-Based Learning Framework</strong><br>
Louis-Martin Rousseau (Polytechnique)</p></li>
<li><p><strong>Unlocking Decision-Making under Uncertainty: The Power of Industry-Academia Collaboration</strong><br>
Emma Frejinger (Université de Montréal), Helen Glover (CN)</p></li>
<li><p><strong>Leveraging Machine Learning and Optimization in Supply Chain Planning</strong><br>
Yossiri Adulyasak (HEC Montréal)</p></li>
</ul>
<p><strong><a href="https://www.gerad.ca/fr/events/2220/view">Full program here</a></strong></p>
<p><img src="/system/assets/000/002/059/2059.horizontalcouleurnoir_card.png " style="border: 0;" >
<img src="/system/assets/000/001/517/1517.Gerad_Logo_CMYK_card.jpg" alt="Logo GERAD" style="border: 0;" >
<img src="/system/assets/000/001/598/1598.ivado-evenement-730x336_card.png" style="border: 0;" ></p>
2024-03-27 08:34:27 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22252024-04-11T11:00:00-04:002024-04-11T11:00:00-04:00On the design of public debate in social networks<p>Séminaire du GERAD</p>
<p>Michel Grabisch, Université Paris 1, France</p>
<p>Chaire de théorie des jeux et gestion<br>
Département de sciences de la décision, HEC Montréal</p>
<p><img src="/system/assets/000/000/744/744.GrabischMichel_card.jpeg" alt="Michel Grabisch"></p>
<p><strong>Salle Trudeau Corporation à HEC Montréal (1<sup>er</sup> étage, section verte).</strong></p>
<p>We propose a model of the joint evolution of opinions and social relationships in a setting where social influence decays over time. The dynamics are based on bounded confidence: social connections between individuals with distant opinions are severed while new connections are formed between individuals with similar opinions. Our model naturally gives raise to strong diversity, i.e., the persistence of heterogeneous opinions in connected societies, a phenomenon that most existing models fail to capture. The intensity of social interactions is the key parameter that governs the dynamics. First, it determines the asymptotic distribution of opinions. In particular, increasing the intensity of social interactions brings society closer to consensus. Second, it determines the risk of polarization, which is shown to increase with the intensity of social interactions. Our results allow to frame the problem of the design of public debates in a formal setting. We hence characterize the optimal strategy for a social planner who controls the intensity of the public debate and thus faces a trade-off between the pursuit of social consensus and the risk of polarization. We also consider applications to political campaigning and show that both minority and majority candidates can have incentives to lead society towards polarization. (with A. Mandel and A. Rusinowska).</p>
2024-03-18 11:47:16 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/21712024-04-10T11:00:00-04:002024-04-10T11:00:00-04:00Route optimization for spring sweeping operations<p>Séminaire “Un(e) chercheur(-euse) du GERAD vous parle!”</p>
<p>Amina Lamghari, Université du Québec à Trois-Rivières, Canada</p>
<p><img src="/system/assets/000/000/890/890.LamghariAmina_card.jpg" alt="Amina Lamghari"></p>
<p>Spring sweeping, the process of cleaning up abrasives (sand and gravel) from roadways on which it was spread in winter, is an important road maintenance practice in countries with severe winters. In this talk, we give an overview of the vehicle routing problem for spring sweeping operations, and we present a model and three decomposition-based approaches for solving it. The efficiency of these approaches is assessed through computational experiments on several classes of both random and real-world instances.</p>
2024-03-25 13:22:15 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22262024-04-09T11:00:00-04:002024-04-09T11:00:00-04:00Fast exact algorithms for some interdiction problems<p>Discussion DS4DM autour d'un café</p>
<p>Ricardo Fukasawa, University of Waterloo, Canada</p>
<p>Chaire d’excellence en recherche du Canada sur la science des données pour la prise de décision en temps réel</p>
<p><img src="/system/assets/000/002/067/2067.RicardoFukasawa_card.jpg" alt="Ricardo Fukasawa"></p>
<p><strong>Séminaire en format hybride au GERAD local 4488 ou <a href="https://hecmontreal.zoom.us/j/81301032985?pwd=WVY0bWFUTHl0ZmQrWmxHZS9IU3ozdz09">Zoom</a>.</strong></p>
<p>While several optimization problems consider that there is a single decision-maker involved, interdiction problems are a form of Stackelberg game where there are two decision-makers involved: a leader and a follower. The follower follows a “classical” optimization problem, trying to optimize a given objective subject to some constraints. The leader is allowed to interdict/block items that are available to the follower, with the goal to make the follower’s objective as bad as possible. This form of zero-sum game can be significantly harder than the original classical counterparts, both in terms of complexity class and in terms of exact algorithm performance.</p>
<p>We study two variants of this problem, with the assumption that the leader must satisfy a knapsack constraint. The different variants relate to the problem the follower solves: knapsack and minimum spanning tree.</p>
<p>The algorithms we develop are based on branch-and-bound with a carefully devised bounding scheme. They improve significantly (by orders of magnitude) the limits of exact algorithms for these problems and show that there is much that can be further researched in this area.</p>
<p>This is joint work with Noah Weninger.</p>
<hr>
<p>Short bio: Ricardo Fukasawa is a Professor at the Department of Combinatorics and Optimization of the University of Waterloo. He did his undergrad and Masters in Electrical Engineering at PUC-Rio, Brazil. He worked at GAPSO Inc developing optimization software for Logistics applications for three years before starting his PhD. He finished his PhD in Algorithms, Combinatorics and Optimization at GeorgiaTech in 2008, under the supervision of Bill Cook. He was the recipient of the 2008-2009 IBM Herman Goldstine postdoctoral fellowship and an Early Research Award from the ministry of colleges and universities in Ontario. He is currently on the editorial board of Mathematical Programming Computation, Operations Research Letters, RAIRO-OR and INFOR. His research interests are in theory and computations for exact algorithms for hard discrete optimization problems. He has works in stochastic programming, bilevel programming, vehicle routing and general mixed-integer programming, as well as works on many applications.</p>
2024-03-26 08:50:46 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22272024-04-05T10:30:00-04:002024-04-05T10:30:00-04:00Webinaire : Geometric Characterization of H-property for Step-graphons<p>Séminaire informel de théorie des systèmes (ISS)</p>
<p>Xudong Chen, Washington University in St. Louis, États-Unis</p>
<p>Centre for intelligent machines (CIM)</p>
<p><img src="/system/assets/000/002/068/2068.Xudong-Chen_card.jpg" alt="Xudong Chen"></p>
<p><strong><a href="https://polymtl-ca.zoom.us/j/84513881004?pwd=VGlOVlZLMlA4Ny95STN1SFRWN0FqUT09">Présentation sur Zoom</a></strong></p>
<p>Graphon has recently been introduced by Lovasz, Sos, etc. to study very large graphs. A graphon can be understood as either the limit object of a convergent sequence of graphs, or, a statistical model from which to sample large random graphs. We take here the latter point of view and address the following problem: What is the probability that a random graph sampled from a graphon has a Hamiltonian decomposition? We have recently observed the following phenomenon: In the asymptotic regime where the size of the random graph goes to infinity, the probability tends to be either 0 or 1, depending on the underlying graphon. In this talk, we establish this “zero-one” property for the class of step-graphons and provide a geometric characterization. </p>
<hr>
<p>Bio: Xudong Chen is an Associate Professor in the Department of Electrical and Systems Engineering at Washington University in St. Louis. He obtained the B.S. degree in Electronic Engineering from Tsinghua University, Beijing, China, in 2009, and the Ph.D. degree in Electrical Engineering from Harvard University, Cambridge, Massachusetts, in 2014. He is an awardee of the 2020 Air Force Young Investigator Program, a recipient of the 2021 NSF Career Award, the recipient of the 2021 Donald P. Eckman Award, and the recipient of the 2023 A.V. Balakrishnan Early Career Award. </p>
<p>His current research interests are in the area of control theory, stochastic processes, optimization, network science, and their applications.</p>
2024-03-25 14:18:52 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22122024-04-04T11:00:00-04:002024-04-04T11:00:00-04:00Webinaire : The social value of information in times of epidemic<p>Séminaire sur les jeux dynamiques et les applications</p>
<p>Chantal Marlats, Université Panthéon-Assas, Paris II, France</p>
<p>Chaire de théorie des jeux et gestion</p>
<p><img src="/system/assets/000/002/037/2037.ChantalMarlats_card.jpeg" alt="Chantal Marlats"></p>
<p><strong><a href="https://hecmontreal.zoom.us/j/82911900343?pwd=ZnpodHo4UFVQQ05Tbi96Q0wrMzFmUT09">Lien pour le webinaire sur Zoom.</a></strong></p>
<p>We analyze an epidemiological model in which individuals trade the costs and benefits of self-isolation while being uncertain about both their type and the dynamics of the epidemic. We characterize the unique symmetric equilibrium and show that uncertainty can be the cause of an additional wave of infections. We calibrate our model to the COVID-19 pandemic and simulate the dynamics of the epidemic under various scenarios to illustrate the impact of uncertainty on self-isolation behaviors. We show that uncertainty about the epidemic dynamics may be welfare improving, both in terms of fraction of deaths and average payoffs. (with Dominique Baril-Tremblay and Lucie Ménager)</p>
2024-01-22 08:47:04 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22092024-03-28T11:00:00-04:002024-03-28T11:00:00-04:00Webinaire : The impacts of environmental policy on industrial allocation: a transboundary pollution dynamic game<p>Séminaire sur les jeux dynamiques et les applications</p>
<p>José R. Morales, Universidad Complutense de Madrid, Espagne</p>
<p>Chaire de théorie des jeux et gestion</p>
<p><img src="/system/assets/000/002/039/2039.JoseAMoralesGarcia_card.jpg" alt="José A. Morales García"></p>
<p><strong><a href="https://hecmontreal.zoom.us/j/82911900343?pwd=ZnpodHo4UFVQQ05Tbi96Q0wrMzFmUT09">Lien pour le webinaire sur Zoom.</a></strong></p>
<p>This paper analyzes a dynamic game between two trading regions that face a transboundary pollution problem. We study how the distribution of firms and trade costs affect the optimal emission policy of governments and how this policy would alter the allocation of the industry. The underlying microeconomic behavior is framed within the Economic Geography literature, in particular within the Footloose Capital Model (FCM). The macroeconomic model that arises is a transboundary pollution linear-quadratic dynamic game. We find that if the damage of pollution is high (low), the region with the larger industrial share reduces (increases) its emissions per firm, and that the steady state pollution reaches a minimum (maximum) when firms are fully concentrated in one region. Additionally, the strategic decisions of governments give rise to a new agglomerative force, absent in the FCM, which could lead to industrial activity fully concentrating in a core region.</p>
2024-01-16 11:34:14 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/21992024-03-27T11:00:00-04:002024-03-27T11:00:00-04:00Webinaire : Modélisation d’événements naturels extrêmes en contexte de climat changeant<p>Séminaire “Un(e) chercheur(-euse) du GERAD vous parle!”</p>
<p>Claudie Ratté-Fortin, HEC Montréal, Canada</p>
<p><img src="/system/assets/000/001/444/1444.RatteFortinClaudie_card.jpg" alt="Claudie Ratté-Fortin"></p>
<p><strong><a href="https://hecmontreal.zoom.us/j/81777698995?pwd=WXplZGdqenhLbGJkSnBKSGtBaFpjdz09">Lien pour le webinaire sur Zoom</a>.</strong></p>
<h5>1. Apprentissage automatique paramétrique vs non-paramétrique pour modéliser la densité de probabilité conditionnelle : application aux blooms de cyanobactéries, Québec, Canada</h5>
<p>Outre l'effet complexe du climat changeant sur les événements naturels extrêmes, leur variabilité spatiotemporelle porte souvent l'empreinte des activités anthropiques sur le territoire. De plus, toute synergie induite par ces activités sur le climat apporte une complexité supplémentaire lors de la modélisation des événements extrêmes. Pour les valeurs extrêmes, une non-stationnarité est souvent apparente sous forme de tendance, possiblement due à un climat changeant à long terme, ou en raison d’une variabilité due à des changements spatiotemporels tels que la physiographie, le climat ou les instruments de mesure utilisés. Dans ce contexte, les modèles paramétriques typiquement utilisés dans la modélisation d’événements extrêmes peuvent devenir inadéquats étant donné la complexité des phénomènes étudiés et leurs changements systématiques dans l'espace et le temps. </p>
<p>L’objectif du projet était d’évaluer l'efficacité des méthodes non-paramétriques d'apprentissage automatique (npML) pour estimer et prédire les valeurs extrêmes associées aux événements naturels. Ces méthodes npML sont comparées à une approche d'apprentissage automatique paramétrique (pML) couramment utilisée, un modèle d'analyse fréquentielle non-stationnaire. Nous avons utilisé une base de données historique compilant la fréquence des algues bleu-vert nuisibles au Québec, Canada. Les résultats montrent qu'un algorithme de densité de probabilité conditionnelle à forêt aléatoire RFCDE à 19 covariables conduit à la meilleure estimation moyenne parmi les modèles considérés. Cependant, pour les quantiles faibles et élevés, le modèle paramétrique RCDE à 4 covariables fournit une meilleure concordance entre les fréquences observées et simulées des efflorescences. Les modèles peuvent être utilisés pour évaluer les effets du changement climatique et des développements anthropogéniques sur la fréquence des HAB. Ils peuvent également servir à mesurer l'efficacité des scénarios d'atténuation et à identifier les zones prioritaires pour les stratégies des plans de restauration.</p>
<h5>2. Accélérer la transition vers des pratiques hivernales durables, intelligentes et connectées</h5>
<p>Bien qu’essentielle au maintien de la sécurité publique, l’application de sel de déglaçage engendre des enjeux environnementaux et économiques majeurs au Canada. Autant pour les administrateurs publics que pour les contracteurs privés, une meilleure gestion devrait passer par une application optimale du sel de déglaçage. Les outils actuellement disponibles afin de déterminer le type et la quantité de sel à épandre sont limités à des tableaux présentant des gammes de quantité à épandre en fonction de gammes de température et d’autres descripteurs météo-routiers. Un rattrapage technologique est nécessaire, d’autant plus que l’interprétation de ces tableaux peut s’avérer complexe et repose souvent sur des décisions subjectives prises au fil du processus de décision. Les pratiques actuelles favorisent ainsi un épandage massif de sel afin d’assurer la sécurité routière des usagers. </p>
<p>L’objectif du projet est de mettre sur pied un projet pilote avec la municipalité de l’Assomption afin de réaliser la preuve de concept d'un dispositif innovant d’aide à la décision (GuiA) permettant l’optimisation de l’épandage de sels de déglaçage et d’abrasifs pour l’entretien hivernal des routes. L’outil innovant basé sur l’intelligence artificielle propose des doses de déglaçants et d’abrasifs en fonction des conditions météo-routières réelles. La sécurité routière sera évaluée au moyen d’un capteur mesurant l’adhérence de la chaussée qui permettra de valider que les recommandations faites par l’outil sont sécuritaires. Un volet de sensibilisation et de vulgarisation sera également mis de l’avant afin d’assurer l’acceptabilité sociale de l’outil IA auprès des citoyens et des cols bleus.</p>
2024-02-14 12:04:54 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22172024-03-19T11:00:00-04:002024-03-19T11:00:00-04:00Neural Heuristics for Mathematical Optimization via Value Function Approximation<p>Discussion DS4DM autour d'un café</p>
<p>Justin Dumouchelle, University of Toronto, Canada</p>
<p>Chaire d’excellence en recherche du Canada sur la science des données pour la prise de décision en temps réel</p>
<p><img src="/system/assets/000/002/064/2064.JustinDumouchelle_card.jpg" alt="Justin Dumouchelle"></p>
<p><strong><a href="https://youtu.be/lQs5l_WB02M">Présentation sur YouTube</a>.</strong></p>
<p>Mathematical optimization is an invaluable framework for modeling and solving decision-making problems with many successes in single-level deterministic problems (e.g., mixed-integer linear or nonlinear optimization). However, many real-world problems require accounting for uncertainty or the reaction of another agent. Paradigms such as stochastic optimization, bilevel optimization, and robust optimization can model these situations but are much slower to solve than their deterministic counterparts, especially when discrete decisions must be made. In this work, we demonstrate how a single learning-based framework, based on value function approximation, can be adapted to all three domains. Empirically, we find solutions of similar, and in some cases significantly better, quality than state-of-the-art algorithms in each field, often within a fraction of the running time. The datasets and three frameworks, Neur2SP (NeurIPS'22), Neur2RO (ICLR'24), and Neur2BiLO (under review at ICML'24), are open-sourced for further research.</p>
2024-03-19 14:01:40 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22142024-03-15T10:30:00-04:002024-03-15T10:30:00-04:00Autonomous learning agents for intelligent neurostimulation<p>Séminaire informel de théorie des systèmes (ISS)</p>
<p>Marco Bonizzato, Polytechnique Montréal, Canada</p>
<p>Centre for intelligent machines (CIM)</p>
<p><img src="/system/assets/000/002/048/2048.bonizzato-marco_card.jpg" alt="Marco Bonizzato"></p>
<p><strong>Séminaire en format hybride au GERAD local 4488 ou <a href="https://polymtl-ca.zoom.us/j/84513881004?pwd=VGlOVlZLMlA4Ny95STN1SFRWN0FqUT09">Zoom</a>.</strong></p>
<p>The nervous system communicates via electrical signals. Electrical neurostimulation is obtained by positioning electrodes in contact with brain, spinal cord or nerves and delivering stimuli that will modulate neuronal activity. This powerful technique allows causal investigation of neural circuits, enabling neuroscientific discovery. It also constitutes the biophysical foundation of a class of medical interventions. Neurostimulation always requires precise adjustment of several stimulation parameters, such as the spatial location of the stimulus, the timing, as well as the frequency of stimulus delivery. Even in the most cutting-edge applications, stimulation tuning has been almost exclusively handled manually. The lack of algorithmic frameworks to control and optimize neurostimulation has hindered scientific discovery. Our program is to transform neurostimulation by introducing an advanced autonomous control layer. We use Gaussian Process-based Bayesian Optimization (GPBO) as an algorithmic framework to tailor and personalize neurostimulation to each individual implant. We show that this framework could be scaled, via algorithmic novelties, to unprecedented neurostimulation steering capacities: 1) from solely stationary to new non-stationary optimization options, 2) from single target to multi-target optimization, 3) from simple outputs to sequences of stimuli.
This work will equip neuroscientists and designers of medical technology with a toolbox of optimization methods to scale the next generation of medical technologies well beyond the limits of the present constrained control.</p>
<hr>
<p>Bio: Marco Bonizzato is an Automation and Life Sciences Engineer working in implantable brain-computer interfaces and neuromodulation technology. He has a double expertise in (a) neural prostheses and (b) machine intelligence and optimization. He is an Assistant Professor of Electrical Engineering at Polytechnique Montréal, Adjunct Professor of Neurosciences at Université de Montréal and Associate Academic Member at Mila - Québec AI institute. He is directing the sciNeurotech Lab. The research goal is developing the entire translational arc of new neurostimulation therapies, aiming at restoring sensorimotor function after neurotrauma, from discovery in rodent to application in human medical technology, tailored and personalized to each user by artificial learning agents. </p>
2024-02-12 08:49:24 -0500GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/21692024-03-13T13:30:00-04:002024-03-13T13:30:00-04:00L’intelligence numérique au service de l’économie circulaire <p>Conférence</p>
<p>CIRODD<br>
GERAD<br>
Réseau de recherche en économie circulaire du Québec (RRECQ) </p>
<h4>Comment l’intelligence numérique peut-elle contribuer à la transition vers l’économie circulaire? Quel rôle jouent les données dans l’atteinte de cet objectif? Le CIRROD, le GERAD et le RRECQ vous invitent à explorer ces questions.</h4>
<p><img src="/system/assets/000/002/027/2027.Symposium RRECQ-CIRODD-GERAD-1200x627-13h30_original.png" style="object-fit: contain; width: 100%; max-width: 100%"><br> </p>
<p>Mettant l’accent sur les données et les outils permettant de guider la transition vers l’économie circulaire, ce symposium gratuit présenté par le CIRROD, le GERAD et le RRECQ s’articulera en quatre temps : deux sessions où chercheuses et chercheurs viendront faire part de leurs travaux, suivies d’un cocktail dînatoire et d’un panel de discussion lors duquel chercheuses et chercheurs, industriels, municipalités et organisations discuteront de leurs préoccupations et de solutions innovantes.</p>
<h3>Programme</h3>
<p>Le programme se décline en deux parties. La première, scindée en deux blocs et composée de présentations scientifiques et d’un cocktail dinatoire pour échanger avec les chercheuses et les chercheurs. La seconde, dédiée à un panel ouvert au public.</p>
<p><strong><a href="https://www.gerad.ca/fr/events/2169/view">Programme court</a></strong></p>
<p><strong><a href="/files/programme_long_2024.pdf">Programme long</a></strong></p>
<details>
<summary>Inscription au panel ouvert au public (18h30 à 20h)</summary>
<p><b><a href="https://airtable.com/appjRYO3Y49ObHDJe/shrIg8NWk13V8Ba3C">S'inscrire au panel seulement.</a></b></p>
<img class="full no-float" src="/system/assets/000/002/047/2047.Symposium RRECQ-CIRODD-GERAD-panel-1080_original.png">
</details>
<details>
<summary>Inscription à la journée complète (13h30 à 20h)</summary>
<p><b><a href="https://symposia.gerad.ca/symposium-economie-circulaire-2024/register">S'inscrire à la journée complète.</a></b></p>
<h3>Première partie : présentations scientifiques et cocktail dinatoire</h3>
<ul>
<li>13 h 30 – Ouverture</li>
<li>13 h 45 à 15 h 30 – <b>Bloc 1</b></li>
<ol>
<li><b>Martin Deron</b> : Le défi numérique : une approche prospective pour la convergence des transitions numériques et écologiques</li>
<li><b>Mir Mostafavi</b> : Vers un jumeau numérique géospatial aux services de l’économie circulaire: le cas de valorisation des matières organiques</li>
<li><b>Amina Lamghari</b> : Transformer les déchets en richesse : comment l’intelligence numérique peut favoriser la circularité dans l’entretien hivernal des routes</li>
<li><b>Christophe Abrassart</b> : Apports et limites des IA génératives pour imaginer des scénarios de transition vers l'économie circulaire</li>
</ol>
<li>15 h 30 à 16 h 00 – Pause</li>
<li>16 h 00 à 17 h 30 – <b>Bloc 2</b></li>
<ol>
<li><b>Jean-Marc Frayret</b> : Optimisation de la conception et du pilotage de systèmes logistiques de l’économie circulaire : projets et défis</li>
<li><b>Samira Keivanpour</b> : L’apport du Machine Learning à la transition vers l’économie circulaire dans le secteur aérospatial</li>
<li><b>Amin Chaabane</b> et <b>Armin Jabbarzadeh</b> : Système de gestion intelligent et durable du marc de café : de déchets à ressources</li>
</ol>
<li>17 h 30 à 18 h 30 – Cocktail dinatoire</li>
</ul>
<h3>Deuxième partie : panel ouvert au public</h3>
<ul>
<li>18 h 30 à 20 h – Panel<br>
Animation : <b>Dominique Anglade</b></li>
</ul>
<img class="full no-float" src="/system/assets/000/002/047/2047.Symposium RRECQ-CIRODD-GERAD-panel-1080_original.png">
</details>
2024-03-19 14:02:30 -0400GERADhttps://www.gerad.ca/tag:www.gerad.ca,2005:Event/22022024-03-13T11:00:00-04:002024-03-13T11:00:00-04:00Strong valid inequalities for a class of concave submodular minimization problems under cardinality constraints<p>Séminaire du GERAD</p>
<p>Qimeng (Kim) Yu, Université de Montréal, Canada</p>
<p><img src="/system/assets/000/002/052/2052.KimYu_card.jpg" alt="Kim Yu"></p>
<p>Applications that involve risk aversion or economies of scale are ubiquitous, such as investment portfolio management and concave cost facility location. In such applications, our task is typically to select items from a given collection to minimize risk/cost, or equivalently, negative utility. We use binary decision variables to represent whether an item is chosen or not, resulting in a discrete optimization problem. Commonly, due to a limited budget, our decision is cardinality-constrained; that is, a pre-specified parameter upper bounds the number of items that we may select. The objective functions that arise from these applications are a class of submodular functions. Finding a tractable convex hull description for the epigraph of any such function under a cardinality constraint has been an open problem. We make significant progress toward tackling this open problem and showcase the practical contributions of our theoretical results on mean-risk optimization, a powerful modeling tool in financial applications. Our experiments demonstrate that our proposed approach leads to significant computational improvement compared to available benchmarks. </p>
<hr>
<p>Bio: Qimeng (Kim) Yu is an assistant professor in the Department of Computer Science and Operations Research (DIRO) at Université de Montréal. She completed her Ph.D. in Industrial Engineering and Management Sciences at Northwestern University in 2023, and she obtained her undergraduate degree in Mathematics at Carleton College in 2018. In her research, she develops theory and algorithms for mixed-integer nonlinear programming to facilitate the solution of complex models with real-world applications. </p>
2024-03-19 14:02:02 -0400GERADhttps://www.gerad.ca/