研究人员为图表示学习（GRL）引入了几种新方法。一种方法“多样性曲线”（Diversity Curves）跟踪图粗化级别的结构多样性，以创建可比较的嵌入。另一种方法“DiGGR”通过学习潜在因子来指导掩码建模，专注于解耦生成图表示学习。此外，“GraphVec”使用谱特征和GIN-Graph Transformer骨干网络将多样化的图向量化为可迁移的嵌入。另一篇论文还提出了“GRL-Safety”，这是一个多轴基准，用于评估GRL方法在各种部署压力下的安全性和可靠性。 AI
影响 图表示学习的进步为分析跨不同领域的复杂关系数据提供了改进的方法。
排序理由 arXiv上发表了多篇关于图表示学习和安全评估基准的新研究论文。
AI 生成摘要 · Google Gemini · 来自 6 个来源。 我们如何撰写摘要 →
arXiv:2605.06466v1 Announce Type: new Abstract: Graph-level representations are crucial tools for characterising structural differences between graphs. However, comparing graphs with different cardinalities, even when sampled from the same underlying distribution, remains challen…
arXiv:2605.06576v1 Announce Type: new Abstract: Graph representation learning (GRL) has evolved from topology-only graph embeddings to task-specific supervised GNNs, and more recently to reusable representations and graph foundation models (GFMs). However, existing evaluations ma…
arXiv:2408.13471v2 Announce Type: replace Abstract: Recently, generative graph models have shown promising results in learning graph representations through self-supervised methods. However, most existing generative graph representation learning (GRL) approaches rely on random ma…
arXiv:2602.04244v2 Announce Type: replace Abstract: Learning universal graph representations across heterogeneous domains is difficult because graph datasets differ in topology, node-attribute semantics, feature dimensions, and even attribute availability. We propose GraphVec, a …
Graph representation learning (GRL) has evolved from topology-only graph embeddings to task-specific supervised GNNs, and more recently to reusable representations and graph foundation models (GFMs). However, existing evaluations mainly measure clean transfer, adaptation, and tas…
Graph-level representations are crucial tools for characterising structural differences between graphs. However, comparing graphs with different cardinalities, even when sampled from the same underlying distribution, remains challenging. Unsupervised tasks in particular require i…