Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Launch Supplement

University of Waterloo

$12,500.00

Assumption-Lean and Adaptive Statistical Learning

EF 2024-25

Résumé
George Box famously wrote that "all models are wrong, some are useful," suggesting we focus efforts on statistical methods which demonstrate utility for decision tasks rather than obsessing over the design of perfect models of the world. The central goal of my research program is to extend this principle to develop theoretical methods that quantify what makes statistical models and assumptions useful even if they are wrong and, from that new knowledge, to develop robust, model-free and assumption-lean decision mechanisms. This is important because, instead of relying on folklore and lucky guesses when choosing among combinations of models and assumptions to base decisions on---and hoping we happen on one "useful" for a given task---we can deploy an assumption-lean, model-free, mechanism and know that we will benefit from the existence of an unknown, simple, "useful" model if one exists. This research program is impactful and timely because of the ubiquity of simplifying models and assumptions in statistics and machine learning, as well as the widespread deployment of machine learning and artificial intelligence tools that lack guarantees on their reliability and robustness. One of the benefits of using simplifying models and assumptions is that they provide optimistic performance guarantees: if they are reasonably representative of the real world relative to their complexity (one notion of "useful"), then we can provide stronger promises for the quality of decisions we make based on those models or assumptions than we could without them. Thus, further motivating the development of assumption-lean decision mechanisms is the principle that truly robust methods must be adaptive, meaning that the methods benefit from the optimistic performance guarantees that simplifying models or assumptions can provide when they are "useful," but remain reliable and as-good-as-possible when simplifying models or assumptions fail, in a way which is agnostic to which setting truly prevails. My research program develops such adaptive, assumption-lean methods for decision making by combining ideas from statistics, computer science, and optimization; particularly adaptive statistical methods and online learning. Specific objectives include the development of statistical learning methods which are robust to the failure of the assumption that data are independent and identically distributed and adapt to the level of that failure, understanding when, why, and how complex models such as neural networks generalize as well as they seem to empirically, and the design of adaptive sequential decision methods. This research program will provide a stimulating environment for HQP that will retain and attract talented trainees in statistics from Canada and abroad. Trainees will develop skills that are highly sought in machine learning research labs in academia, and in industry at tech companies, health and pharmaceutical companies, online retailers, financial firms, and others.
Programme
Discovery Launch Supplement
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
DGECR-2024-00455
Name nom
Negrea, Jeffrey
Department département
Statistics and Actuarial Science
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
12500
Program id
DGECR
Program name en
Discovery Launch Supplement
Program name fr
Supplément Tremplin vers la découverte
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1205
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Mathematical sciences
Area of application fr
Sciences mathématiques
Research subject code
3001
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Statistical theory
Research subject fr
Théorie de statistique
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
Assumption-Lean and Adaptive Statistical Learning
Keywords
statistical learning; robustness; adaptivity; machine learning; sequential decision making; sequential prediciton; online learning; assumption-lean; optimization; bandits
Application summary
George Box famously wrote that "all models are wrong, some are useful," suggesting we focus efforts on statistical methods which demonstrate utility for decision tasks rather than obsessing over the design of perfect models of the world. The central goal of my research program is to extend this principle to develop theoretical methods that quantify what makes statistical models and assumptions useful even if they are wrong and, from that new knowledge, to develop robust, model-free and assumption-lean decision mechanisms. This is important because, instead of relying on folklore and lucky guesses when choosing among combinations of models and assumptions to base decisions on---and hoping we happen on one "useful" for a given task---we can deploy an assumption-lean, model-free, mechanism and know that we will benefit from the existence of an unknown, simple, "useful" model if one exists. This research program is impactful and timely because of the ubiquity of simplifying models and assumptions in statistics and machine learning, as well as the widespread deployment of machine learning and artificial intelligence tools that lack guarantees on their reliability and robustness. One of the benefits of using simplifying models and assumptions is that they provide optimistic performance guarantees: if they are reasonably representative of the real world relative to their complexity (one notion of "useful"), then we can provide stronger promises for the quality of decisions we make based on those models or assumptions than we could without them. Thus, further motivating the development of assumption-lean decision mechanisms is the principle that truly robust methods must be adaptive, meaning that the methods benefit from the optimistic performance guarantees that simplifying models or assumptions can provide when they are "useful," but remain reliable and as-good-as-possible when simplifying models or assumptions fail, in a way which is agnostic to which setting truly prevails. My research program develops such adaptive, assumption-lean methods for decision making by combining ideas from statistics, computer science, and optimization; particularly adaptive statistical methods and online learning. Specific objectives include the development of statistical learning methods which are robust to the failure of the assumption that data are independent and identically distributed and adapt to the level of that failure, understanding when, why, and how complex models such as neural networks generalize as well as they seem to empirically, and the design of adaptive sequential decision methods. This research program will provide a stimulating environment for HQP that will retain and attract talented trainees in statistics from Canada and abroad. Trainees will develop skills that are highly sought in machine learning research labs in academia, and in industry at tech companies, health and pharmaceutical companies, online retailers, financial firms, and others.