Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

Simon Fraser University

$62,000.00

Leveraging the Power of Information Networks: Uniting Logical Rules, Graphical Models, and Graph Representation Learning

FY 2024-25

Summary
Objectives. Information networks capture information about different types of entities and different types of links. Links can be material as in a computer network, or represent abstract relationships, like friendships in a social network. Heterogeneous information networks (aka knowledge graphs) are ubiquitous, maintained by many organizations in a relational database. Deep graph learning techniques have achieved remarkable success on tasks such as link-based classification, clustering, and graph classification. The goal of my research program is to advance neural graph learning by incorporating domain knowledge. The representation of knowledge must be machine-readable and easy to interpret and modify by the user. My plan is to use first-order Bayesian networks, which marry graphical models for visualizing statistical associations with the expressive power of symbolic logic through highly interpretable probabilistic rules (e.g. "if a person works in a city, they are also likely to live in that city"). In previous research supported by NSERC, my group has developed a model discovery system known as FactorBase, that takes as input a heterogeneous information network and outputs a first-order Bayesian network. FactorBase supports a human-in-the-loop mixed initiative approach where users can easily specify their domain knowledge. My research program will investigate how the symbolic knowledge represented in a Bayesian network can enhance learning a deep generative graph model, which is the most powerful type of deep graph neural network. Combining symbolic knowledge and neural methods is one of the current frontiers of AI known as neuro-symbolic AI. I propose a moment-matching approach to leverage rule-based knowledge for neuro-symbolic AI: Training a generative model to match expected and observed rule instantiation counts. Significance. The benefits of this project for building deep graph generative models are as follows. 1. Generating more realistic graphs. For instance, the example rule will ensure that generated graphs exhibit a correlation between working and living in the same city. 2. Increased predictive power. Better graph generative models support more accurate predictions on downstream predictive tasks such as link-based classification, clustering, and graph classification. For instance, the example rule allows us to infer, with some confidence, that a person lives in a city if they work there. 3. Interpretability and user control. User control, together with the transparency and interpretability of Bayesian networks, enhances the user's trust. The utility of rule-based graph generation will be evaluated against graph benchmark datasets, and an application project in traffic modelling: Trajectory prediction in an autonomous vehicle (AV) traffic scenario. My project will explore how learning rules about traffic behaviour can enhance traffic interaction models that include interactions among traffic participants.
Program
Discovery Grants Program - Individual
Location
British Columbia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2024-05924
Name nom
Schulte, Oliver
Department département
Computing Science
Organization id
5
Institution établissement
Simon Fraser University
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
62000
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
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
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
Leveraging the Power of Information Networks: Uniting Logical Rules, Graphical Models, and Graph Representation Learning
Keywords
graph generative models; graph representation learning; first-order logical rules; statistical-relational learning; neural-symbolic artificial intelligence
Application summary
Objectives. Information networks capture information about different types of entities and different types of links. Links can be material as in a computer network, or represent abstract relationships, like friendships in a social network. Heterogeneous information networks (aka knowledge graphs) are ubiquitous, maintained by many organizations in a relational database. Deep graph learning techniques have achieved remarkable success on tasks such as link-based classification, clustering, and graph classification. The goal of my research program is to advance neural graph learning by incorporating domain knowledge. The representation of knowledge must be machine-readable and easy to interpret and modify by the user. My plan is to use first-order Bayesian networks, which marry graphical models for visualizing statistical associations with the expressive power of symbolic logic through highly interpretable probabilistic rules (e.g. "if a person works in a city, they are also likely to live in that city"). In previous research supported by NSERC, my group has developed a model discovery system known as FactorBase, that takes as input a heterogeneous information network and outputs a first-order Bayesian network. FactorBase supports a human-in-the-loop mixed initiative approach where users can easily specify their domain knowledge. My research program will investigate how the symbolic knowledge represented in a Bayesian network can enhance learning a deep generative graph model, which is the most powerful type of deep graph neural network. Combining symbolic knowledge and neural methods is one of the current frontiers of AI known as neuro-symbolic AI. I propose a moment-matching approach to leverage rule-based knowledge for neuro-symbolic AI: Training a generative model to match expected and observed rule instantiation counts. Significance. The benefits of this project for building deep graph generative models are as follows. 1. Generating more realistic graphs. For instance, the example rule will ensure that generated graphs exhibit a correlation between working and living in the same city. 2. Increased predictive power. Better graph generative models support more accurate predictions on downstream predictive tasks such as link-based classification, clustering, and graph classification. For instance, the example rule allows us to infer, with some confidence, that a person lives in a city if they work there. 3. Interpretability and user control. User control, together with the transparency and interpretability of Bayesian networks, enhances the user's trust. The utility of rule-based graph generation will be evaluated against graph benchmark datasets, and an application project in traffic modelling: Trajectory prediction in an autonomous vehicle (AV) traffic scenario. My project will explore how learning rules about traffic behaviour can enhance traffic interaction models that include interactions among traffic participants.
Simon Fraser University - $62,000 Grant (2024-25) | Canada Spends