Open archwalker opened 5 years ago
Hello, it looks like that step1 is not right if I understand it correctly. In step1, for the node 4, it gets (1, 14) not (1, 4).
@h12345jack Hello, it looks like that step1 is not right if I understand it correctly. In step1, for the node 4, it gets (1, 14) not (1, 4).
node index after comma are all neighbor nodes' indices, thus 1 is excluded.
@archwalker
@h12345jack Hello, it looks like that step1 is not right if I understand it correctly. In step1, for the node 4, it gets (1, 14) not (1, 4).
node index after comma are all neighbor nodes' indices, thus 1 is excluded.
I am sorry, I mean if some nodes are not labeled correctly.
@h12345jack
@archwalker
@h12345jack Hello, it looks like that step1 is not right if I understand it correctly. In step1, for the node 4, it gets (1, 14) not (1, 4).
node index after comma are all neighbor nodes' indices, thus 1 is excluded.
I am sorry, I mean if some nodes are not labeled correctly.
you are right, I have fix the problem, thanks
请问为什么后话部分说“单层感知机不是单射函数”呀?单层感知机由矩阵乘法、激活函数组成,矩阵乘法是单调函数,激活函数如果也是单调的(比如tanh、sigmoid),那么由于单调函数一定是单射函数,单层感知机就是单设函数啊?
@liangchen1ceeee 请问为什么后话部分说“单层感知机不是单射函数”呀?单层感知机由矩阵乘法、激活函数组成,矩阵乘法是单调函数,激活函数如果也是单调的(比如tanh、sigmoid),那么由于单调函数一定是单射函数,单层感知机就是单设函数啊?
您好,矩阵乘法不一定是单射的,矩阵乘法单射的充要条件是矩阵为列满秩矩阵。
https://archwalker.github.io/blog/2019/06/22/GNN-Theory-WL.html
一、引言前面的文章中,我们介绍了GNN的三个基本模型GCN、GraphSAGE、GAT,分析了经典的GCN逐层传播公式是如何由谱图卷积推导而来的。GNN模型现在正处于学术研究的热点话题,那么我们不经想问,GNN模型到底有多强呢?我们的目的是分析GNN的表达能力,我们需要一个模型作为衡量标准。比如说如果我们想衡量G...