Fig. 5From: Gumbel-softmax-based optimization: a simple general framework for optimization problems on graphsResults on hyper-parameters tuning of population size \(N_{{\text{bs}}}\), evolution cycle T, and substitution ratio 1/u on SK model with 1024 and 8192 nodes. Experiment configuration: initial \(\tau = 20\), final \(\tau = 1\), and learning rate \(\eta\) = 1. The results are averaged for 1250 instances with 1024 nodes and 100 instances with 8192 nodes, respectivelyBack to article page