[论文精读]Variational Graph Auto-Encoders

夏莉莉iy

于 2024-07-01 19:55:23 发布

阅读量969

点赞数 17

分类专栏：论文精读文章标签：人工智能深度学习机器学习笔记神经网络贴图算法

本文链接：https://blog.csdn.net/Sherlily/article/details/140107856

版权

论文精读专栏收录该内容

67 篇文章 8 订阅

订阅专栏

论文网址：[1611.07308] Variational Graph Auto-Encoders (arxiv.org)

英文是纯手打的！论文原文的summarizing and paraphrasing。可能会出现难以避免的拼写错误和语法错误，若有发现欢迎评论指正！文章偏向于笔记，谨慎食用

2.1. A latent variable model for graph-structured data

2.2. Experiments on link prediction

3. Reference

1. 省流版

1.1. 心得

（1）好短的文章捏，只有两页

1.2. 论文总结图

2. 论文逐段精读

2.1. A latent variable model for graph-structured data

①Task: unsupervised learning

②Latent space of unsupervised VGAE in Cora, a citation network dataset:

③Definitions: for undirected and unweighted graph $G=\left ( V,E \right )$ , the number of nodes $N=\left | V \right |$ , the adjacency matrix with self-loop and the diagnal elements all set to 1defined as $A$ , the degree matrix is $\mathbf{D}$ , the stochastic latent variables is $z_i \in \mathbb{R}^{1 \times F}$ , $\mathbf{Z}=\left [ z_1,z_2,...,z_N \right ] \in\mathbb{R}^{N \times F}$ , node feature matrix $\mathbf{X} \in \mathbb{R}^{N \times D}$ （但是没说这个节点特征是啥，估计自己随便定义吧）

④Inference model:

$q(\mathbf{Z}\mid\mathbf{X},\mathbf{A})=\prod_{i=1}^Nq(\mathbf{z}_i\mid\mathbf{X},\mathbf{A})$

with $q(\mathbf{z}_i\mid\mathbf{X},\mathbf{A})=\mathcal{N}(\mathbf{z}_i\mid\boldsymbol{\mu}_i,\mathrm{diag}(\boldsymbol{\sigma}_i^2))$

where $\boldsymbol\mu = \mathrm{GCN}_{\boldsymbol{\mu}}(\mathbf{X},\mathbf{A})$ is the matrix of mean vectors $\mu _i$ ;

$\log \boldsymbol \sigma = \mathrm{GCN}_{\boldsymbol{\sigma}}(\mathbf{X},\mathbf{A})$ （为啥左边要有个log啊）

⑤A 2 layer GCN:

$\mathrm{GCN}(\mathbf{X},\mathbf{A})=\mathbf{\tilde{A}}\mathrm{ReLU}(\mathbf{\tilde{A}}\mathbf{X}\mathbf{W}_{0})\mathbf{W}_{1}$

where $\mathbf{W}_{i}$ denotes weight matrix, $\mathbf{\tilde{A}}=\mathbf{D}^{-\frac{1}{2}}\mathbf{A}\mathbf{D}^{-\frac{1}{2}}$

⑥ $\mathrm{GCN}_{\boldsymbol{\mu}}(\mathbf{X},\mathbf{A})$ 和 $\mathrm{GCN}_{\boldsymbol{\sigma}}(\mathbf{X},\mathbf{A})$ 共享 $\mathbf{W}_{0}$ 的参数？？？什么玩意儿？？为啥有俩，是引用了之前的什么高斯吗？

⑦Generative model:

$p\left(\mathbf{A}\mid\mathbf{Z}\right)=\prod_{i=1}^{N}\prod_{j=1}^{N}p\left(A_{ij}\mid\mathbf{z}_{i},\mathbf{z}_{j}\right)$

with $p\left(A_{ij}=1 | \mathbf{z}_i,\mathbf{z}_j\right)=\sigma(\mathbf{z}_i^\top\mathbf{z}_j)$

where $\sigma \left ( \cdot \right )$ represents the logistic sigmoid function

⑧Loss function:

$\mathcal{L}=\mathbb{E}_{q(\mathbf{Z}|\mathbf{X},\mathbf{A})}\big[\log p\left(\mathbf{A}\left|\mathbf{Z}\right)\right]-\mathrm{KL}\big[q(\mathbf{Z}\left|\mathbf{X},\mathbf{A}\right)\|p(\mathbf{Z})\big]$