Abstract
Graph learning focuses on predicting labels or properties of graph-structured data, such as social networks, molecular structures, and citation networks. Unlike traditional data formats, graphs encode both node-level features and edge-based relationships, which makes them complex yet rich sources of information. Traditional graph neural networks typically emphasize individual graph representations, neglecting the relational context that could help with graph learning. In this dissertation, we introduce novel frameworks for graph learning that address limitations in existing graph neural network methods by exploring relationships among graphs, graph augmentations, and intricate relational patterns captured within graphs. We provide new perspectives on graph learning, with implications for applications in games, power systems, and beyond.