Empowering Heterogeneous Graph Foundation Models via Decoupled Relation Alignment
2026-05-04T08:52:15Z•17855e30b940bc1b6c7346a7c916d7e551d6676a1a8d818751f8ca5766776de6
Aitchison embeddingsDRSAFMAFairness Monitor AgentGRE-MCLLM-generated codeSocialBias-Benchbenchmarkingcompositional embeddingsfairness in code generationfeature-structure decouplingfew-shot transfergraph foundation modelsgraph retrievalgraph transformerheterogeneous graphsmodality completionmulti-agent pipelinesmulti-domain graphsmultimodal recommendationplug-and-play preprocessingrelation alignmentsimplex representations','ILR coordinatessocial biassparse-routing codebook
What happened
This silver document is an arXiv RSS snapshot containing multiple CS/ML papers (May 4, 2026). Key contributions: (1) "Decoupled Relation Subspace Alignment (DRSA)" — a plug-and-play relation-driven preprocessing for Graph Foundation Models on multi-domain heterogeneous graphs that decouples feature semantics from relation structure via dual-relation subspace projection and a feature-structure decomposition; demonstrates improved cross-domain and few-shot transfer and provides code. (2) "Social Bias in LLM-Generated Code" — a large empirical benchmark (SocialBias-Bench, 343 tasks) showing high,
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_si
- Record identifier
- 17855e30b940bc1b6c7346a7c916d7e551d6676a1a8d818751f8ca5766776de6
- Enrichment time
- 2026-05-04T08:52:15Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.