In figure 8a, we compare solution qualities of degree centrality (DC), POMCP and PSINET-S, PSINET-W, and PSINET-C on BTER networks of varying sizes.
In figure 8b, we show run times of DC, POMCP, and PSINET-S, PSINET-W, and PSINET-C on the same BTER networks.
In figure 9a, we show solution quality of PSINET-W with increasing network instances, for a (40, 71, 41) BTER network with a horizon of 10.
In figure 9b, we show run time of PSINET-W with increasing network instances, for a (40, 71, 41) BTER network with a horizon of 10.
Figure 10a shows the solution quality, when PSINET-W and DC are solved with different p(e) values on the network edges (the p(e) values were changed for both the network that was input to the algorithm and the network on which the algorithm's policy was executed) for a (40, 71, 41) BTER network with a horizon of 10.
In figure 10b, we show solution qualities of PSINET-W and DC on a (30, 31, 27) BTER network by varying the number of nodes selected per round (k).
We also use a small test set for weak scaling experiments, consisting of 2D grids and BTER graphs.
We examine how the local greedy cycles perform as graph size increases with weak scaling experiments on 2D grid graphs and BTER graphs.
The weak scaling experiments shown in Figure 10 show that, with the exception of a BTER outlier, the work scales nicely with graph size.
Tarek Tantawi (BTER Foundation Research Fellow), Lynn Wang (BTER Foundation Communications Intern), and Katherine Watt (BTER Foundation Research Intern).
BioTherapeutics, Education & Research (BTER) Foundation, 36 Urey Court, Irvine, CA 92617, USA