Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

Université de Montréal

$36,000.00

Machine Learning in the Presence of Model-Dependent Data

FY 2024-25

Summary
Machine learning (ML) has revolutionized our approach to computer vision, natural language processing, and recommendation algorithms. In addition, we expect ML to significantly remodel many fields, such as drug discovery and healthcare. By curating large high-quality datasets, machine learning researchers have established well-defined milestones. On the one hand, such tasks with a fixed dataset provide an excellent way for the community to measure progress and foster innovation. They are undoubtedly one of the critical elements of the recent successes of ML. On the other hand, the community now needs to move beyond these relatively simple tasks as they do not fully represent the deployment at scale of ML models in the real world. To make machine learning models more efficient and reliable for real-world applications, we need to address the fact that, in many situations, the behavior of the users-who generate the data-hinges on the large-scale deployment of the models themselves. With our recognized expertise in adversarial formulation and learning in games, this research program will advance the field of ML in the presence of model-dependent data via the following directions: 1) Most deep learning models are vulnerable to imperceptible changes in the input distribution. Understanding and fixing this lack of robustness is critical for many applications and, more generally, as a path toward a better understanding of deep neural networks. We will develop new techniques to learn and measure robust representation. 2)In many situations (recommendation system, pricing), the deployed machine learning models will influence the user's choices, thus inducing changes in the data distribution. For example, recommending a specific route to too many users may create a traffic jam and make this recommendation suboptimal. We will provide novel tools to understand performativity with deep neural networks. 3) ML systems often interact with other strategic agents when deployed in the real world. The latter can be machine learning systems or humans competing (or collaborating) in the same environment. We will advance the field of variational inequality optimization by proposing and analyzing state-of-the-art methods. Undertaking this problem of model-dependent data is critical for many industrial actors whose developed heuristics lack a principled understanding of the problem. Breakthrough progress on this topic will have a direct impact on many existing machine learning models already deployed in the real world. Notably, most of the machine learning algorithms deployed on social media platforms, which affect billions of people, lack robustness and are performative. With its high potential impact, this research program will contribute to maintaining Canada's leadership position in artificial intelligence.
Program
Discovery Grants Program - Individual
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2023-04373
Name nom
Gidel, Gauthier
Department département
DIRO
Organization id
63
Institution établissement
Université de Montréal
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
36000
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
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
1207
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Information, computer and communication technologies
Area of application fr
Technologie de l'information, des ordinateurs et des réseaux de communications
Research subject code
2805
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Learning and inference theories
Research subject fr
Théories de l'apprentissage et l'inférence
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
Machine Learning in the Presence of Model-Dependent Data
Keywords
machine learning; performative prediction; adversarial robustness; multi-player game optimization; variational inequality problem; adversarial training; game theory; optimization; large scale
Application summary
Machine learning (ML) has revolutionized our approach to computer vision, natural language processing, and recommendation algorithms. In addition, we expect ML to significantly remodel many fields, such as drug discovery and healthcare. By curating large high-quality datasets, machine learning researchers have established well-defined milestones. On the one hand, such tasks with a fixed dataset provide an excellent way for the community to measure progress and foster innovation. They are undoubtedly one of the critical elements of the recent successes of ML. On the other hand, the community now needs to move beyond these relatively simple tasks as they do not fully represent the deployment at scale of ML models in the real world. To make machine learning models more efficient and reliable for real-world applications, we need to address the fact that, in many situations, the behavior of the users-who generate the data-hinges on the large-scale deployment of the models themselves. With our recognized expertise in adversarial formulation and learning in games, this research program will advance the field of ML in the presence of model-dependent data via the following directions: 1) Most deep learning models are vulnerable to imperceptible changes in the input distribution. Understanding and fixing this lack of robustness is critical for many applications and, more generally, as a path toward a better understanding of deep neural networks. We will develop new techniques to learn and measure robust representation. 2)In many situations (recommendation system, pricing), the deployed machine learning models will influence the user's choices, thus inducing changes in the data distribution. For example, recommending a specific route to too many users may create a traffic jam and make this recommendation suboptimal. We will provide novel tools to understand performativity with deep neural networks. 3) ML systems often interact with other strategic agents when deployed in the real world. The latter can be machine learning systems or humans competing (or collaborating) in the same environment. We will advance the field of variational inequality optimization by proposing and analyzing state-of-the-art methods. Undertaking this problem of model-dependent data is critical for many industrial actors whose developed heuristics lack a principled understanding of the problem. Breakthrough progress on this topic will have a direct impact on many existing machine learning models already deployed in the real world. Notably, most of the machine learning algorithms deployed on social media platforms, which affect billions of people, lack robustness and are performative. With its high potential impact, this research program will contribute to maintaining Canada's leadership position in artificial intelligence.