notesum.ai

Published at November 18

The GECo algorithm for Graph Neural Networks Explanation

cs.AI

Released Date: November 18, 2024

Authors: Salvatore Calderaro¹, Domenico Amato¹, Giosuè Lo Bosco¹, Riccardo Rizzo², Filippo Vella²

Aff.: ¹Dept of Mathematics and Computer Science, University of Palermo, Palermo, 90123, Italy; ²Institute for High Performance Computing and Networking, National Research Council of Italy, Palermo, 90146, Italy

Arxiv: http://arxiv.org/abs/2411.11391v1

Dataset	Method	$Fid^{+}$ $\uparrow$	$Fid^{-}$ $\downarrow$	charact $\uparrow$	GEA $\uparrow$
ba_house_cycle	random	$0.411\pm 0.067$	$0.410\pm 0.053$	$0.477\pm 0.055$	$0.145\pm 0.004$
	PGMExplainer	$0.270\pm 0.056$	$0.436\pm 0.048$	$0.360\pm 0.048$	$0.089\pm 0.006$
	PGExplainer	$0.217\pm 0.137$	$0.459\pm 0.101$	$0.292\pm 0.152$	$0.213\pm 0.152$
	GNNExplainer	$0.478\pm 0.044$	$0.257\pm 0.066$	$0.579\pm 0.037$	$0.190\pm 0.001$
	SubgraphX	$0.191\pm 0.181$	$0.356\pm 0.252$	$0.270\pm 0.233$	$0.269\pm 0.147$
	GECo	$\textbf{0.929}\pm\textbf{0.043}$	$\textbf{0.000}\pm\textbf{0.002}$	$\textbf{0.952}\pm\textbf{0.024}$	$\textbf{0.305}\pm\textbf{0.029}$
ba_cycle_wheel	random	$0.297\pm 0.033$	$0.294\pm 0.034$	$0.417\pm 0.035$	$0.172\pm 0.005$
	PGMExplainer	$0.133\pm 0.024$	$0.360\pm 0.040$	$0.219\pm 0.034$	$0.113\pm 0.007$
	PGExplainer	$0.111\pm 0.193$	$0.424\pm 0.133$	$0.139\pm 0.237$	$0.155\pm 0.149$
	GNNExplainer	$0.491\pm 0.035$	$0.026\pm 0.018$	$0.652\pm 0.032$	$0.234\pm 0.003$
	SubgraphX	$0.329\pm 0.231$	$0.218\pm 0.300$	$0.439\pm 0.306$	$0.380\pm 0.181$
	GECo	$\textbf{0.607}\pm\textbf{0.100}$	$\textbf{0.000}\pm\textbf{0.000}$	$\textbf{0.750}\pm\textbf{0.076}$	$\textbf{0.553}\pm\textbf{0.032}$
er_house_cycle	random	$0.368\pm 0.041$	$0.372\pm 0.040$	$0.461\pm 0.030$	$0.154\pm 0.005$
	PGMExplainer	$0.203\pm 0.053$	$0.430\pm 0.040$	$0.295\pm 0.058$	$0.097\pm 0.006$
	PGExplainer	$0.286\pm 0.154$	$0.426\pm 0.141$	$0.367\pm 0.169$	$0.265\pm 0.174$
	GNNExplainer	$0.471\pm 0.039$	$0.084\pm 0.030$	$0.621\pm 0.035$	$0.197\pm 0.021$
	SubgraphX	$0.260\pm 0.111$	$0.349\pm 0.110$	$0.363\pm 0.132$	$0.262\pm 0.130$
	GECo	$\textbf{0.791}\pm\textbf{0.090}$	$\textbf{0.000}\pm\textbf{0.001}$	$\textbf{0.880}\pm\textbf{0.057}$	$\textbf{0.391}\pm\textbf{0.069}$
er_cycle_wheel	random	$0.348\pm 0.055$	$0.352\pm 0.058$	$0.448\pm 0.038$	$0.170\pm 0.005$
	PGMExplainer	$0.213\pm 0.057$	$0.442\pm 0.046$	$0.303\pm 0.056$	$0.113\pm 0.005$
	PGExplainer	$0.222\pm 0.164$	$0.414\pm 0.134$	$0.295\pm 0.190$	$0.227\pm 0.167$
	GNNExplainer	$0.454\pm 0.044$	$0.048\pm 0.036$	$0.613\pm 0.040$	$0.227\pm 0.003$
	SubgraphX	$0.167\pm 0.124$	$0.392\pm 0.133$	$0.243\pm 0.160$	$0.263\pm 0.135$
	GECo	$\textbf{0.866}\pm\textbf{0.108}$	$\textbf{0.002}\pm\textbf{0.018}$	$\textbf{0.923}\pm\textbf{0.070}$	$\textbf{0.407}\pm\textbf{0.050}$
ba_cycle_wheel_grid	random	$0.467\pm 0.034$	$0.468\pm 0.036$	$0.495\pm 0.023$	$0.185\pm 0.004$
	PGMExplainer	$0.229\pm 0.035$	$0.127\pm 0.005$	$0.299\pm 0.030$	$0.127\pm 0.005$
	PGExplainer	$0.214\pm 0.172$	$0.568\pm 0.099$	$0.260\pm 0.168$	$0.159\pm 0.114$
	GNNExplainer	$0.664\pm 0.032$	$0.147\pm 0.038$	$0.746\pm 0.026$	$0.256\pm 0.003$
	SubgraphX	$0.562\pm 0.142$	$0.217\pm 0.197$	$0.650\pm 0.161$	$0.527\pm 0.115$
	GECo	$\textbf{0.887}\pm\textbf{0.052}$	$\textbf{0.000}\pm\textbf{0.002}$	$\textbf{0.939}\pm\textbf{0.029}$	$\textbf{0.561}\pm\textbf{0.036}$
er_cycle_wheel_grid	random	$0.520\pm 0.039$	$0.521\pm 0.034$	$0.496\pm 0.016$	$0.185\pm 0.004$
	PGMExplainer	$0.331\pm 0.051$	$0.623\pm 0.036$	$0.349\pm 0.032$	$0.128\pm 0.005$
	PGExplainer	$0.285\pm 0.153$	$0.557\pm 0.135$	$0.333\pm 0.151$	$0.205\pm 0.143$
	GNNExplainer	$0.628\pm 0.041$	$0.074\pm 0.024$	$0.748\pm 0.033$	$0.246\pm 0.002$
	SubgraphX	$0.325\pm 0.087$	$0.410\pm 0.095$	$0.415\pm 0.089$	$0.378\pm 0.098$
	GECo	$\textbf{0.915}\pm\textbf{0.028}$	$\textbf{0.001}\pm\textbf{0.000}$	$\textbf{0.954}\pm\textbf{0.017}$	$\textbf{0.510}\pm\textbf{0.038}$