CHoE: Cross-Domain Heterogeneous Graph Prompt Learning via Structure-Conditioned Experts
Researchers have developed CHoE, a novel method for Heterogeneous Graph Prompt Learning (HGPL) designed to overcome limitations in cross-domain applications. Unlike existing methods that are primarily in-domain, CHoE utilizes a network of structure-conditioned experts. This approach enables better performance in few-shot cross-domain scenarios by employing a structure-aware routing mechanism to select compatible experts and integrating representations across multiple views for prediction. Experiments demonstrate CHoE's consistent improvement over baseline methods in cross-domain few-shot learning. AI
IMPACT Enhances few-shot learning capabilities for heterogeneous graph data across different domains.