【论文分享】小样本半监督图结点分类模型 Meta-PN：Meta Propagation Networks for Graph Few-shot Semi-supervised Learning

• 题目：Meta Propagation Networks for Graph Few-shot Semi-supervised Learning
• 链接：https://arxiv.org/abs/2112.09810
• 源码：https://github.com/kaize0409/Meta-PN
• 会议：AAAI Conference on Artificial Intelligence (AAAI) 2022. CCF-A
• 时间：2022-04-01
• 摘要：本文提出了一种新的图元学习框架：元传播网络(Meta- PN)，用于解决小样本半监督节点分类问题。该框架由两个部分构成 1. Adaptive Label Propagator (Meta Learner)：利用目标模型的反馈来调整其传播策略，以在未标记节点上推断出准确的伪标签。2. Feature-label Transformer (Target Model)： 吸收伪标记节点的结构和特征知识预测标签，从而解决了小样本半监督学习背后的挑战。
  

  目录

  • 0 PPT
  • 1 Introduction
  • 2 Method
  • • Architecture
    • • Adaptive Label Propagator (Meta Learner)
      • Feature-label Transformer (Target Model)
      • Learning to Propagate
    • Model Learning via Bi-level Optimization
  • 3 Experiment
  • • Dataset
    • Compared Methods
    • Implementation Details
    • Evaluation Results
  • 4 Conclusion

0 PPT