Gumbel-softmax-based optimization: a simple general framework for optimization problems on graphs

Computational Social Networks

Table 6 Results on influence maximization problems compared to classic methods

Graph	Size	Algorithm	Number of initial nodes
Graph	Size	Algorithm	1	2	3	4	5
Karate	34	Greedy	7.573	\({13.632}\)*	\({18.438}\)*	\({21.320}\)*	\({23.446}\)*
		SA	7.573	13.770	18.467	21.843	24.205
		GSO^a	7.573	13.770	18.467	21.843	24.205
Jazz	198	Greedy	80.283	109.707	125.922	135.396	143.803
		SA	80.283	109.707	125.922	135.396	143.803
		GSO^a	80.283	109.707	125.922	135.396	143.803
Email	1133	Greedy	19.392	34.859	50.322	65.774	81.137
		SA	19.392	34.859	50.322	65.774	81.137
		GSO^a	19.392	34.859	\({50.317}\)*	\({65.515}\)*	\({80.576}\)*
Government	7057	Greedy	4070.021	4507.093	4725.925	4869.137	5008.638
		SA	4070.020	4296.233	4331.295	4375.324	4452.940
		GSO^a	4070.021	\({4433.256}\)*	\({4639.054}\)*	\({4788.753}\)*	\({4905.063}\)*

The best and the second best results are denoted in italic and asterisk, respectively
^aConfiguration: batch size = 128, fixed \(\tau\) = 1, learning rate = 1, \(\alpha\) = 5, instance = 10, and epoch = 2000