This paper introduces the concept of Data-Efficient Graph Learning (DEGL) and provides a comprehensive survey of recent advancements in this field. The authors highlight the challenges associated with training models on large labeled datasets, particularly in low-resource settings, and explore three key areas: self-supervised graph learning, semi-supervised graph learning, and few-shot graph learning. Each area is further divided into subcategories, such as generation-based, contrastive-based, and auxiliary property-based methods for self-supervised learning, and label propagation, consistency regularization, and pseudo-labeling for semi-supervised learning. The paper also discusses the application of DEGL in various domains, including molecular structures, genomics, and recommendation systems. Additionally, it addresses the challenges and future directions in DEGL, emphasizing the need for more robust and versatile models that can handle out-of-distribution data and improve explainability. The authors conclude by suggesting potential areas for future research, such as combining large language models with graph learning and developing data-efficient non-Euclidean GNN models.This paper introduces the concept of Data-Efficient Graph Learning (DEGL) and provides a comprehensive survey of recent advancements in this field. The authors highlight the challenges associated with training models on large labeled datasets, particularly in low-resource settings, and explore three key areas: self-supervised graph learning, semi-supervised graph learning, and few-shot graph learning. Each area is further divided into subcategories, such as generation-based, contrastive-based, and auxiliary property-based methods for self-supervised learning, and label propagation, consistency regularization, and pseudo-labeling for semi-supervised learning. The paper also discusses the application of DEGL in various domains, including molecular structures, genomics, and recommendation systems. Additionally, it addresses the challenges and future directions in DEGL, emphasizing the need for more robust and versatile models that can handle out-of-distribution data and improve explainability. The authors conclude by suggesting potential areas for future research, such as combining large language models with graph learning and developing data-efficient non-Euclidean GNN models.