The paper presents Dual-Prism (DP), a new method for graph data augmentation that addresses issues like graph property distortions and restricted structural changes. DP, comprising DP-Noise and DP-Mask, preserves the critical properties of graphs while generating augmented versions. These techniques focus on maintaining low-frequency eigenvalues unchanged, which has been found to conserve important graph properties. The paper suggests that DP represents a promising new direction for graph data augmentation, with experimental results validating its effectiveness.
Publication date: 18 Jan 2024
Project Page: https://arxiv.org/abs/2401.09953v1
Paper: https://arxiv.org/pdf/2401.09953
