How Powerful are Graph Neural Networks?
[Blog : Paper Notes sorted out by paper]
🍭 About This Paper
Keyulu Xu, Weihua Hu, Jure Leskovec, Stefanie Jegelka
MIT and Stanford University
https://arxiv.org/pdf/1810.00826.pdf
🍭 GNNs
一个图神经网络的更新准则一般可以归纳为这样:
a v ( k ) = A G G R E G A T E ( k ) ( { h u ( k − 1 ) : u ∈ N ( v ) } ) , h v ( k ) = COMBINE ( k ) ( h v ( k − 1 ) , a v ( k ) ) a_{v}^{(k)}= AGGREGATE ^{(k)}\left(\left\{h_{u}^{(k-1)}: u \in \mathcal{N}(v)\right\}\right), \quad h_{v}^{(k)}=\operatorname{COMBINE}^{(k)}\left(h_{v}^{(k-1)}, a_{v}^{(k)}\right) av(k)=AGGREGATE(k)({ hu(k−1):u∈N(v)}),hv(k)=COMBINE(k)(hv(k−1),av(k