Back to search

University of Crete (UOC)

Combining structured knowledge and data-driven representations for object states classification

Abstract

dc:description

This dissertation investigates the task of Object State Classification (OSC) addressing the challenges of recognizing object states in images. We integrate symbolic knowledge from Knowledge Graphs (KGs) with data-driven Graph Neural Networks (GNNs) to enhance OSC performance and generalization capabilities, particularly in zero-shot learning scenarios. Our core contribution is a novel approach to Object-agnostic Zero-Shot State Classification that leverages KGs to encode commonsense knowledge and employs GNNs to learn expressive embeddings capturing rich semantic information. Crucially, our method operates without relying on explicit object recognition, increasing robustness to appearance variations and applicability in scenarios where object recognition may be unreliable. Through extensive experiments on four benchmark datasets, we demonstrate our method's effectiveness in recognizing unseen state classes, achieving significant improvements over baseline approaches and state-of-the-art zero-shot learning techniques. We further investigate the use of Large Language Models (LLMs) to enhance KG construction and utilization for OSC, exploring their capacity to generate semantically rich corpora for training visual models and to construct more informative knowledge representations. Additionally, we explore metapath learning to capture complex semantic relationships in KGs, enhancing both performance and interpretability of the proposed method. In conclusion, this dissertation advances OSC through the development of novel methodologies that synergistically combine KGs, GNNs, and LLMs, the creation of new datasets, and the introduction of innovative research objectives. Our findings underscore the potential of neurosymbolic AI integration for achieving more robust, interpretable, and accurate OSC methods, laying the groundwork for future research in this domain.

Degree

thesis:*
Grantor dc:publisher
University of Crete (UOC)
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Gouidis, Filippos
  • Γουΐδης, Φίλιππος

Subjects

dc:subject × 20

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/58987

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Gouidis, Filippos; Γουΐδης, Φίλιππος. Combining structured knowledge and data-driven representations for object states classification. University of Crete (UOC), 2025. http://hdl.handle.net/10442/hedi/58987