TABLE 2.

For each combination of a deep learning model, a learning rate schedule, a test type, and an optimization method presented in Sections IV-C to IV-E, the summary statistics of the average estimation error across the five runs.

Deep learning model Learning rate schedule		Test type	Optimization method Average estimation error
GoogLeNet	fixed learning rate	unloaded system test	Adam	0.272±0.060
	fixed learning rate	unloaded system test	RMSprop	0.417±0.140
	fixed learning rate	workload interference test	Adam	0.400±0.074
	exponential decay	unloaded system test	Adam	1.033±0.169
	step decay	unloaded system test	Adam	0.372±0.028
GRU	fixed learning rate	unloaded system test	Adam	0.362±0.034
	exponential decay	unloaded system test	Adam	0.303±0.074