Table 7:
Comparison of the error rates (%), the computational times, and the average numbers of epochs for network training, on the CIFAR-10 dataset with the different parameter $N$. We run each method ten times and report the errors in the format of “best (mean $±$ std).” The values of the computational time and the average number of epochs are the mean over the ten runs.
MethodError rateTime (days)Average # epochs
CGP-CNN (ConvSet) $5.92(6.48±0.48)$ 15.6 50.0
with early termination ($N=3$$5.61(6.52±0.59)$ 3.32 8.97
with early termination ($N=5$$5.81(6.34±0.39)$ 6.73 18.8
with early termination ($N=10$$6.40(6.93±0.50)$ 11.4 30.6
CGP-CNN (ResSet) $5.01(6.10±0.89)$ 14.7 50.0
with early termination ($N=3$$5.30(6.22±0.46)$ 2.39 7.96
with early termination ($N=5$$5.42(6.34±0.61)$ 4.93 15.1
with early termination ($N=10$$5.27(6.12±0.50)$ 6.05 23.9
CGP-CNN (ResSet) with RichInit $4.90(5.60±0.52)$ 11.7 50.0
with early termination ($N=3$$5.06(5.83±0.47)$ 1.29 6.88
with early termination ($N=5$$5.07(5.76±0.57)$ 2.86 12.4
with early termination ($N=10$$4.90(5.54±0.60)$ 6.36 22.2
MethodError rateTime (days)Average # epochs
CGP-CNN (ConvSet) $5.92(6.48±0.48)$ 15.6 50.0
with early termination ($N=3$$5.61(6.52±0.59)$ 3.32 8.97
with early termination ($N=5$$5.81(6.34±0.39)$ 6.73 18.8
with early termination ($N=10$$6.40(6.93±0.50)$ 11.4 30.6
CGP-CNN (ResSet) $5.01(6.10±0.89)$ 14.7 50.0
with early termination ($N=3$$5.30(6.22±0.46)$ 2.39 7.96
with early termination ($N=5$$5.42(6.34±0.61)$ 4.93 15.1
with early termination ($N=10$$5.27(6.12±0.50)$ 6.05 23.9
CGP-CNN (ResSet) with RichInit $4.90(5.60±0.52)$ 11.7 50.0
with early termination ($N=3$$5.06(5.83±0.47)$ 1.29 6.88
with early termination ($N=5$$5.07(5.76±0.57)$ 2.86 12.4
with early termination ($N=10$$4.90(5.54±0.60)$ 6.36 22.2
Close Modal