ARTFEED — Contemporary Art Intelligence

VISION Model Advances Graph Few-Shot Learning with In-Context Learning

other · 2026-05-26

A new model named VISION (adVancIng graph few-Shot learning via In-cOntext LearNing) has been proposed to address key limitations in graph few-shot learning. The approach reframes the problem as a fine-tuning-free sequence reasoning task, inspired by in-context learning in large language models. VISION uses a context-aware network with role embeddings and a dual-context fusion module to integrate unlabeled nodes without complex task adaptation. The research is detailed in arXiv preprint 2605.24410.

Key facts

  • VISION is a model for graph few-shot learning
  • It uses in-context learning inspired by large language models
  • The approach is fine-tuning-free
  • It leverages unlabeled nodes in graphs
  • The model uses role embeddings and a dual-context fusion module
  • The research is published on arXiv with ID 2605.24410
  • Graph few-shot learning aims to classify nodes from novel classes with few labeled examples
  • Existing methods often rely on supervised tasks and require complex adaptation

Entities

Institutions

  • arXiv

Sources