Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

The University of British Columbia

$52,000.00

Foundations of Learning and Reasoning about Entities and Relations under Uncertainty

FY 2023-24

Summary
With environmental and other data coming online at an unprecedented rate through government action and mandates in many parts of the world, it is past time to build predictive models for such relational data that can match the predictive models of text and vision. We are a long way from being able to add an environmental dataset into a learner and get reasonable answers. This proposal is about the foundations of making (probabilistic) predictions about relational data, which has been studied in areas such as statistical relational AI, relational learning, and knowledge graph learning. We recently completed a research monograph on statistical relational AI and an edited book on lifted inference, which exploits relational structure in probabilistic reasoning.      My long-term objective is to develop the foundations of relational reasoning and decision making. This includes building useful tools for real-world domains as well as understanding the limitations and the potential of various methods. The short and medium term objectives of this proposal concentrate on two open problems: missing data and aggregation.      Relational data sets typically only contain positive information. All of the false and nearly all of all true values are missing. Missing data cannot be ignored unless it it missing completely at random, which is a strong assumption that does not usually arise. Principled methods for handling missing data involve conditioning on the fact that data is missing. This is infeasible in relational models, as there are combinatorially many possible statements. We will investigate building models assuming everything is missing, and then update it on what isn't missing.      The second problem is that of aggregation, for example predicting the age of a person from the movies they have rated, or predicting sources of pollution from upstream and downstream measurements of various chemicals. How relational probabilistic models handle aggregation is built in to their semantics (typically equivalent to logistic regression or noisy-or). Graph neural networks allow for a choice of a fixed set of aggregators (e.g., mean, sum , max, attention). All of the suggestions have undesirable properties, including asymptotic behavior as the number of related entities approaches infinity (perhaps in different orders). We will investigate novel models of aggregation that work better in theory and practice.     The past work has been done with students at all levels. I work in a very hands-on approach with a diverse group of students. I cater to each student's interests and skills, ranging from purely theoretic to very applied. Many of my students work with local industry on real-world problems. My graduating students have gone onto jobs with local and international companies that match their skills learned.
Program
Discovery Grants Program - Individual
Location
British Columbia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2023-04497
Name nom
Poole, David
Department département
Computer Science
Organization id
2
Institution établissement
The University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2023
Competition year année de concours
2023
Award amount
52000
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
1200
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Advancement of knowledge
Area of application fr
Avancement des connaissances
Research subject code
2805
Research subject group en
Artificial intelligence
Research subject group fr
Learning and inference theories
Research subject en
Intelligence artificielle
Research subject fr
Théories de l'apprentissage et l'inférence
Application title
Foundations of Learning and Reasoning about Entities and Relations under Uncertainty
Keyword
aggregation; knowledge representation; missing data; reasoning under uncertainty; relational learning; statistical relational artificial intelligence
Application summary
With environmental and other data coming online at an unprecedented rate through government action and mandates in many parts of the world, it is past time to build predictive models for such relational data that can match the predictive models of text and vision. We are a long way from being able to add an environmental dataset into a learner and get reasonable answers. This proposal is about the foundations of making (probabilistic) predictions about relational data, which has been studied in areas such as statistical relational AI, relational learning, and knowledge graph learning. We recently completed a research monograph on statistical relational AI and an edited book on lifted inference, which exploits relational structure in probabilistic reasoning.      My long-term objective is to develop the foundations of relational reasoning and decision making. This includes building useful tools for real-world domains as well as understanding the limitations and the potential of various methods. The short and medium term objectives of this proposal concentrate on two open problems: missing data and aggregation.      Relational data sets typically only contain positive information. All of the false and nearly all of all true values are missing. Missing data cannot be ignored unless it it missing completely at random, which is a strong assumption that does not usually arise. Principled methods for handling missing data involve conditioning on the fact that data is missing. This is infeasible in relational models, as there are combinatorially many possible statements. We will investigate building models assuming everything is missing, and then update it on what isn't missing.      The second problem is that of aggregation, for example predicting the age of a person from the movies they have rated, or predicting sources of pollution from upstream and downstream measurements of various chemicals. How relational probabilistic models handle aggregation is built in to their semantics (typically equivalent to logistic regression or noisy-or). Graph neural networks allow for a choice of a fixed set of aggregators (e.g., mean, sum , max, attention). All of the suggestions have undesirable properties, including asymptotic behavior as the number of related entities approaches infinity (perhaps in different orders). We will investigate novel models of aggregation that work better in theory and practice.     The past work has been done with students at all levels. I work in a very hands-on approach with a diverse group of students. I cater to each student's interests and skills, ranging from purely theoretic to very applied. Many of my students work with local industry on real-world problems. My graduating students have gone onto jobs with local and international companies that match their skills learned.
The University of British Columbia - $52,000 Grant (2… | Canada Spends