Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

University of Waterloo

$38,000.00

Inverse Learning for Decision Making Under Uncertainty

FY 2023-24

Summary
The abundance of data in recent years provides an invaluable opportunity to learn from past experiences. By processing data on historical decisions that were made and the outcomes that followed, whether favourable or unfavourable, practitioners can potentially learn to make better decisions that are more likely to result in desirable outcomes in the future. While there is considerable momentum to learn from data through artificial intelligence, machine learning, and statistics, the field of Operations Research (OR) has not been using this valuable resource to its full potential. An underutilized, yet mathematically rigorous tool in OR that can benefit from this abundance of data is inverse optimization. Traditional (forward) optimization formulates a system as a mathematical model and generates optimal decisions. Inverse optimization works oppositely by observing past decisions and inferring the underlying mathematical model that resulted in such a decision. Inverse learning is the process of using inverse and forward optimization models iteratively to learn from past data and improve future decisions. Traditional inverse optimization has largely focused on inputting deterministic observations that are assumed to be optimal. The literature almost exclusively focuses on learning objective functions in linear settings. Due to the inherent complexity of inverse optimization, inferring more realistic complex models, non-optimal data, and noise and uncertainty in data have not been well studied in the literature. This program focuses on inverse learning of complex models in data-driven applications where noise and uncertainty are inevitable. The proposed methodology accounts for suboptimal input and will also learn from unacceptable decisions that resulted in unfavourable outcomes. The inverse optimization models will consider different complex underlying systems and propose solution methods that overcome the computational complexity of solving such models. Using our built expertise in robust optimization, uncertainty mitigation techniques will be embedded into the proposed inverse learning framework which will allow us to deal with real-world data. An application of the proposed methodologies will be demonstrated in radiation therapy treatment planning for cancer. While certain clinical guidelines exist on how radiation beams must be designed for different cancer types, these guidelines are not universally agreed upon and the implicit success criteria are not exactly known. Using historical data on radiation beam setup for a large pool of retrospective patients, the inverse framework can learn the success criteria for different treatment modalities, which have underlying models of varying complexities. In contrast to commonly used black-box learning models, our inverse models have a clear mathematical structure, and can therefore provide high-quality treatments that are transparent, interpretable, and verifiable by clinical experts.
Program
Discovery Grants Program - Individual
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2023-03802
Name nom
Mahmoudzadeh, Houra
Department département
Management Science and Engineering
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2023
Competition year année de concours
2023
Award amount
38000
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
1100
Area of application group en
Health, education and social services
Area of application group fr
Santé publique, éducation et services sociaux
Area of application en
Health, education and social services
Area of application fr
Santé publique, éducation et services sociaux
Research subject code
2950
Research subject group en
Applied mathematics
Research subject group fr
Applied mathematics
Research subject en
Mathématiques appliquées
Research subject fr
Mathématiques appliquées
Application title
Inverse Learning for Decision Making Under Uncertainty
Keyword
Inference; Integer Programming; Inverse Optimization; Learning; Linear Programming; Mathematical Optimization; Operations Research; Robust Optimization
Application summary
The abundance of data in recent years provides an invaluable opportunity to learn from past experiences. By processing data on historical decisions that were made and the outcomes that followed, whether favourable or unfavourable, practitioners can potentially learn to make better decisions that are more likely to result in desirable outcomes in the future. While there is considerable momentum to learn from data through artificial intelligence, machine learning, and statistics, the field of Operations Research (OR) has not been using this valuable resource to its full potential. An underutilized, yet mathematically rigorous tool in OR that can benefit from this abundance of data is inverse optimization. Traditional (forward) optimization formulates a system as a mathematical model and generates optimal decisions. Inverse optimization works oppositely by observing past decisions and inferring the underlying mathematical model that resulted in such a decision. Inverse learning is the process of using inverse and forward optimization models iteratively to learn from past data and improve future decisions. Traditional inverse optimization has largely focused on inputting deterministic observations that are assumed to be optimal. The literature almost exclusively focuses on learning objective functions in linear settings. Due to the inherent complexity of inverse optimization, inferring more realistic complex models, non-optimal data, and noise and uncertainty in data have not been well studied in the literature. This program focuses on inverse learning of complex models in data-driven applications where noise and uncertainty are inevitable. The proposed methodology accounts for suboptimal input and will also learn from unacceptable decisions that resulted in unfavourable outcomes. The inverse optimization models will consider different complex underlying systems and propose solution methods that overcome the computational complexity of solving such models. Using our built expertise in robust optimization, uncertainty mitigation techniques will be embedded into the proposed inverse learning framework which will allow us to deal with real-world data. An application of the proposed methodologies will be demonstrated in radiation therapy treatment planning for cancer. While certain clinical guidelines exist on how radiation beams must be designed for different cancer types, these guidelines are not universally agreed upon and the implicit success criteria are not exactly known. Using historical data on radiation beam setup for a large pool of retrospective patients, the inverse framework can learn the success criteria for different treatment modalities, which have underlying models of varying complexities. In contrast to commonly used black-box learning models, our inverse models have a clear mathematical structure, and can therefore provide high-quality treatments that are transparent, interpretable, and verifiable by clinical experts.