Structured Knowledge Representation for Reasoning in Deep Models

Huang, Sirui

Structured Knowledge Representation for Reasoning in Deep Models

Huang, Sirui

Permalink

Publication Type:: Thesis
Issue Date:: 2025

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (17.85 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Huang, Sirui
dc.date.accessioned	2026-04-08T08:13:00Z
dc.date.available	2026-04-08T08:13:00Z
dc.date.issued	2025
dc.identifier.uri	http://hdl.handle.net/10453/194612
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Structured knowledge—represented through tabular formats, bipartite graphs, and knowledge graphs—pervades real-world systems as a foundational organizational framework for critical decision-making processes in domains such as laboratory documentation, financial reporting. These structures enable explicit organization of knowledge while encapsulating insights through patterns and relationships unattainable via unstructured data alone. Despite their operational value, the construction of such knowledge structures remains resource-intensive, necessitating specialized domain expertise and enterprise-level system integrations. Consequently, optimizing the utilization of these meticulously organized knowledge repositories becomes imperative. While deep learning demonstrates proficiency in raw data processing, its capacity for symbolic reasoning over structured knowledge remains constrained by inherent challenges: reconciling discrete relational logic with neural architectures, addressing sparse connectivity, and maintaining interpretability. This thesis systematically addresses these limitations through a unified methodological framework that advances reasoning capabilities across structured knowledge systems, encompassing innovations from data-level infrastructure to representation-level architectures. At the data-level, we establish a unified benchmarking protocol to examine distinctive structural properties while enabling comprehensive analytical capabilities. From representation-level, we engineer three novel computational models that resolve critical challenges including static heterogeneous dependency management and dynamic adaptation mechanisms within structured knowledge systems. By innovatively synthesizing human-curated knowledge with data-driven learning paradigms, this research advances artificial intelligence systems capable of harnessing structured knowledge's full potential, thereby mediating the historical dichotomy between symbolic reasoning and neural learning approaches.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/194612/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	©2025 Sirui Huang
dc.rights	au.edu.uts.lib/cph
dc.title	Structured Knowledge Representation for Reasoning in Deep Models	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Structured knowledge—represented through tabular formats, bipartite graphs, and knowledge graphs—pervades real-world systems as a foundational organizational framework for critical decision-making processes in domains such as laboratory documentation, financial reporting. These structures enable explicit organization of knowledge while encapsulating insights through patterns and relationships unattainable via unstructured data alone. Despite their operational value, the construction of such knowledge structures remains resource-intensive, necessitating specialized domain expertise and enterprise-level system integrations. Consequently, optimizing the utilization of these meticulously organized knowledge repositories becomes imperative. While deep learning demonstrates proficiency in raw data processing, its capacity for symbolic reasoning over structured knowledge remains constrained by inherent challenges: reconciling discrete relational logic with neural architectures, addressing sparse connectivity, and maintaining interpretability. This thesis systematically addresses these limitations through a unified methodological framework that advances reasoning capabilities across structured knowledge systems, encompassing innovations from data-level infrastructure to representation-level architectures. At the data-level, we establish a unified benchmarking protocol to examine distinctive structural properties while enabling comprehensive analytical capabilities. From representation-level, we engineer three novel computational models that resolve critical challenges including static heterogeneous dependency management and dynamic adaptation mechanisms within structured knowledge systems. By innovatively synthesizing human-curated knowledge with data-driven learning paradigms, this research advances artificial intelligence systems capable of harnessing structured knowledge's full potential, thereby mediating the historical dichotomy between symbolic reasoning and neural learning approaches.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/194612