Centrala begrepp
Addressing out-of-distribution generalization challenges in heterogeneous graph few-shot learning through a causal model.
Sammanfattning
The article introduces the COHF model to handle distribution shifts in heterogeneous graphs, focusing on OOD generalization. It discusses the challenges and proposes solutions using a structural causal model. The methodology includes a variational autoencoder-based HGNN and explores the invariance principle for OOD generalization.
- Introduction to Heterogeneous Graphs and Label Sparsity Issue
- Existing Methods' Assumptions and Challenges Faced in Real-world Scenarios
- Novel Problem of Out-of-Distribution (OOD) Generalization in HGFL
- Multi-level and Phase-spanning Characteristics of OOD Environments in HGFL
- Importance of Consistency Across Source HG, Training Data, and Testing Data
- Proposed COHF Model and Key Contributions
- Related Work on Heterogeneous Graph Representation Learning, Graph Few-shot Learning, OOD Generalization on Graphs, and Domain-Invariant Feature Learning
Statistik
"Extensive experiments on seven real-world datasets have demonstrated the superior performance of COHF over the state-of-the-art methods."
"Our work is the first to propose the novel problem of OOD generalization in heterogeneous graph few-shot learning."
Citat
"Our key contributions are proposing the COHF model for causal modeling in HGFL."
"We conduct extensive experiments demonstrating COHF's superiority over existing methods."