Graph data augmentation has emerged as a critical area of research, particularly for improving generalization and robustness in graph neural networks (GNNs). Among various approaches, Input Mixup, which generates virtual samples by interpolating input features and labels, has gained significant attention in domains such as image classification and natural language processing for its simplicity and effectiveness. However, adapting Mixup to node classification presents unique challenges due to irregularity in graph structures and the difficulty of aligning and interpolating neighboring nodes. This seminar will explore recent advancements in graph data augmentation, focusing on the iGraphMix method designed specifically for node classification tasks. iGraphMix addresses the irregularity and alignment issues by generating virtual nodes and their edges through feature and label interpolation while sampling neighboring nodes. These virtual graphs enhance GNN training by serving as augmented inputs, offering compatibility with diverse GNN architectures and augmentations. We will discuss the theoretical foundations of iGraphMix, which demonstrate its potential to improve generalization performance, as well as empirical results validating these claims. Additionally, we will delve into the limitations of iGraphMix in heterophilic graph settings, where its performance can degrade, and highlight emerging research efforts to overcome these challenges. Beyond theory, this seminar will showcase real-world applications where graph data augmentation methods have been successfully deployed, such as stock-price prediction and social network analysis. This session aims to provide theoretical insights, practical perspectives, and application-driven case studies, fostering discussions on advancing data-centric methodologies for graph deep learning.