Scholarlink

<p>Graph foundation models and graph prompts have emerged as a popular research area for graph learning research in recent years. However, their real-world deployment raises concerns about the potential leakage of sensitive information during inference. This work presents a new graph encryption for private graph foundation model (GFM) inference with applications for sparse graphs. It adds dummy edges into graphs randomly and encrypts graph structures rather than adjacency matrices through homomorphic encryption (HE). Based on this kind of graph encryption, we propose a new private GFM inference scheme with graph prompts, and the basic idea is to perform private neighbor queries for the multiplications of adjacency matrices, and integrate neighbor sampling to reduce computation costs. Our scheme achieves a computational complexity of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({\cal{O}}(M)\)</EquationSource> <EquationSource Format="MATHML"><math display="block"> <mrow> <mrow> <mi mathvariant="script">O</mi> </mrow> </mrow> <mo stretchy="false">(</mo> <mi>M</mi> <mo stretchy="false">)</mo> </math></EquationSource> </InlineEquation> for HE multiplications, which is a remarkable improvement over the previous <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\({\cal{O}}(N)^{2}\)</EquationSource> <EquationSource Format="MATHML"><math display="block"> <mrow> <mrow> <mi mathvariant="script">O</mi> </mrow> </mrow> <mo stretchy="false">(</mo> <mi>N</mi> <msup> <mo stretchy="false">)</mo> <mrow> <mn>2</mn> </mrow> </msup> </math></EquationSource> </InlineEquation> complexity. Here, <i>N</i> and <i>M</i> are the numbers of nodes and edges in a graph, respectively. Theoretically, we prove the correctness of our private prompted GFM inference and the security of our encryption scheme. Finally, we have conducted extensive experiments to validate the effectiveness and efficiency of our private GFM inference.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

On the encryption for graph foundation model inference of sparse graph

  • 期刊论文

  • 发表日期：

    2025年5月22日

查看全文
  • Man-Jie Yuan,
  • Xue-Tong Bai,
  • Kai-Ran Zhang,
  • Wei Gao

摘要

Graph foundation models and graph prompts have emerged as a popular research area for graph learning research in recent years. However, their real-world deployment raises concerns about the potential leakage of sensitive information during inference. This work presents a new graph encryption for private graph foundation model (GFM) inference with applications for sparse graphs. It adds dummy edges into graphs randomly and encrypts graph structures rather than adjacency matrices through homomorphic encryption (HE). Based on this kind of graph encryption, we propose a new private GFM inference scheme with graph prompts, and the basic idea is to perform private neighbor queries for the multiplications of adjacency matrices, and integrate neighbor sampling to reduce computation costs. Our scheme achieves a computational complexity of \({\cal{O}}(M)\) O ( M ) for HE multiplications, which is a remarkable improvement over the previous \({\cal{O}}(N)^{2}\) O ( N ) 2 complexity. Here, N and M are the numbers of nodes and edges in a graph, respectively. Theoretically, we prove the correctness of our private prompted GFM inference and the security of our encryption scheme. Finally, we have conducted extensive experiments to validate the effectiveness and efficiency of our private GFM inference.