Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

École Polytechnique de Montréal

$50,000.00

Efficient derivative-free optimization methods for transportation systems and machine learning

EF 2024-25

Résumé
This proposal describes a five-year research program to develop new methods for computationally expensive BlackBox Optimization (BBO). BBO refers to situations in which the structure of the optimization problem (i.e., the objective function and the constraints) is unknown and cannot be exploited. Practical software implementations will be proposed to solve challenging problems related to transportation systems and machine learning. This proposal is divided into two parts, each part concerns several projects that are well-suited for students. Derivative-Free Optimization (DFO) and Bayesian Optimization (BO) have been successfully applied to solve BBO problems. Despite addressing the same class of problems, only few interactions between DFO and BO were established. In fact, recent research made significant progress on algorithmic and theoretical aspects for DFO methods. On the other side, although BO is seen as a powerful strategy to solve computationally expensive BBO problems, it offers limited convergence guarantees. In the first part of this proposal, I aim to reduce the gap between DFO and BO by designing efficient methods for high dimensional computationally expensive settings, particularly in settings where the BBO problem is expensive to evaluate, constrained, and potentially noisy. I also plan to improve the efficiency of state-of-the-art DFO methods. The second part of this proposal concerns a structured class of BBO problems, known as bilevel, where two nested optimization problems are involved. Bilevel BBO has received extensive attention thanks to the recent advances in machine learning and engineering design where many challenging optimization problems naturally possess a bilevel structure and are computationally expensive BBO. For bilevel optimization, the majority of existing research is derivative based and there is only limited work for bilevel BBO, mainly by using either heuristic-based techniques or by imposing stringent assumptions. In this context, I aim to develop practical, efficient and convergent DFO strategies to solve bilevel BBO problems that are likely expensive-to-evaluate and subject to stochastic uncertainties. I also plan in this part to study a specific subclass of bilevel problems which is given in a min-max optimization form. I aim to use high-level information about the objectives, the constraints and the nature of the problem (when available) to enhance the efficiency of the proposed DFO methods. This proposal will contribute to the training of 10 graduate and undergraduate students who will receive a multidisciplinary training and will be involved in international collaborations. They will develop advanced skills in optimization, computer science, machine learning, urban transportation, air traffic, and aircraft design. It will also lead to developments that are of high interest for many industries in Canada (e.g., Bombardier and Hydro-Québec) and abroad (e.g., Google and Airbus).
Programme
Discovery Grants Program - Individual
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2024-05093
Name nom
Diouane, Youssef
Department département
Mathematics and Industrial Engineering Department
Organization id
64
Institution établissement
École Polytechnique de Montréal
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
50000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1509
Committee name en
Civil, Industrial and Systems Engineering
Committee name fr
Génie civil, industriel et des systèmes
Area of application code
1206
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Engineering
Area of application fr
Génie
Research subject code
1601
Research subject group en
Industrial engineering
Research subject group fr
Génie industriel
Research subject en
Operations research and management science
Research subject fr
Recherche opérationnelle et science de la gestion
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Efficient derivative-free optimization methods for transportation systems and machine learning
Keywords
numerical optimization; derivative-free optimization; bayesian optimization; software developments; engineering applications; transportation systems; machine learning; aircraft design
Application summary
This proposal describes a five-year research program to develop new methods for computationally expensive BlackBox Optimization (BBO). BBO refers to situations in which the structure of the optimization problem (i.e., the objective function and the constraints) is unknown and cannot be exploited. Practical software implementations will be proposed to solve challenging problems related to transportation systems and machine learning. This proposal is divided into two parts, each part concerns several projects that are well-suited for students. Derivative-Free Optimization (DFO) and Bayesian Optimization (BO) have been successfully applied to solve BBO problems. Despite addressing the same class of problems, only few interactions between DFO and BO were established. In fact, recent research made significant progress on algorithmic and theoretical aspects for DFO methods. On the other side, although BO is seen as a powerful strategy to solve computationally expensive BBO problems, it offers limited convergence guarantees. In the first part of this proposal, I aim to reduce the gap between DFO and BO by designing efficient methods for high dimensional computationally expensive settings, particularly in settings where the BBO problem is expensive to evaluate, constrained, and potentially noisy. I also plan to improve the efficiency of state-of-the-art DFO methods. The second part of this proposal concerns a structured class of BBO problems, known as bilevel, where two nested optimization problems are involved. Bilevel BBO has received extensive attention thanks to the recent advances in machine learning and engineering design where many challenging optimization problems naturally possess a bilevel structure and are computationally expensive BBO. For bilevel optimization, the majority of existing research is derivative based and there is only limited work for bilevel BBO, mainly by using either heuristic-based techniques or by imposing stringent assumptions. In this context, I aim to develop practical, efficient and convergent DFO strategies to solve bilevel BBO problems that are likely expensive-to-evaluate and subject to stochastic uncertainties. I also plan in this part to study a specific subclass of bilevel problems which is given in a min-max optimization form. I aim to use high-level information about the objectives, the constraints and the nature of the problem (when available) to enhance the efficiency of the proposed DFO methods. This proposal will contribute to the training of 10 graduate and undergraduate students who will receive a multidisciplinary training and will be involved in international collaborations. They will develop advanced skills in optimization, computer science, machine learning, urban transportation, air traffic, and aircraft design. It will also lead to developments that are of high interest for many industries in Canada (e.g., Bombardier and Hydro-Québec) and abroad (e.g., Google and Airbus).