1
Introduction
spectral graph theory
graph signal processing (GSP)
Fourier basis
eigenvalue decomposition (EVD)
2
Proposed Technique
Generalizing CNNs to graphs requires three fundamental steps:
(i) the design of localized convolutional fifilters on graphs
(ii) a graph coarsening procedure that groups together similar vertices
(iii) a graph pooling operation that trades spatial resolution for higher fifilter resolution.
2.1
Learning Fast Localized Spectral Filters
There are two strategies to defifine convolutional fifilters; either from a spatial approach or from a spectral approach.
Graph Fourier Transform
![](https://img-blog.csdnimg.cn/20200602112848443.png)
![](https://img-blog.csdnimg.cn/20200602112934546.png)
![](https://img-blog.csdnimg.cn/20200602112952200.png)
![](https://img-blog.csdnimg.cn/20200602113027448.png)
A signal
x : V → R defined on the nodes of the graph may be regarded as a vector
![](https://img-blog.csdnimg.cn/20200602113228868.png)
x
i
is the value of
x
at the
i
th node
,
,
(这个D也就是每个点有几条边,加权)
L标准化定义:
![](https://img-blog.csdnimg.cn/20200602113815752.png)
![](https://img-blog.csdnimg.cn/20200602113851558.png)
As L is a real symmetric positive semidefifinite matrix(实对称半正定矩阵
),it has a complete set of orthonormal eigenvectors
(一组完全正交的特征向量),known as the graph Fourier modes, and their associated ordered real nonnegative eigenvalues
(相关的有序实非负特征值),
identified as the frequencies of the graph.
![](https://img-blog.csdnimg.cn/20200602114200965.png)
![](https://img-blog.csdnimg.cn/20200602114339745.png)
L分解:
,
,
![](https://img-blog.csdnimg.cn/20200602225356610.png)
![](https://img-blog.csdnimg.cn/20200602225416289.png)
![](https://img-blog.csdnimg.cn/20200602225438458.png)
所以:把一个信号x的图傅里叶变换定义为:
![](https://img-blog.csdnimg.cn/20200602225608249.png)
反过来:
![](https://img-blog.csdnimg.cn/20200602225651885.png)
Spectral fifiltering of graph signals
![](https://img-blog.csdnimg.cn/20200602232449382.png)
![](https://img-blog.csdnimg.cn/20200602232504722.png)
![](https://img-blog.csdnimg.cn/2020060223252254.png)
![](https://img-blog.csdnimg.cn/20200602233242991.png)
Polynomial parametrization for localized fifilters
non-parametric fifilters两个局限:
(i) they are not localized in space
(ii) their learning complexity is in
O
(
n
), the dimensionality of the data.
所以,用下面这个
polynomial fifilter
![](https://img-blog.csdnimg.cn/2020060223313973.png)
![](https://img-blog.csdnimg.cn/20200602233319422.png)
(可能内容不全)