論文筆記：Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering

前言 初代頻域GCN簡單粗暴的將 d i a g ( g ^ ( λ l ) ) diag(\hat{g}{(\lambda_l)}) diag(g^​(λl​))變成了卷積核 d i a g ( θ l ) diag(\theta_l) diag(θl​) 由此最終形式爲 y o u t p u t = σ ( U g θ ( Λ ) U T x ) y_{output}=\sigma(U{g_

>>阅读原文<<