Local Epochs Inefficiency Caused by Device Heterogeneity in Federated Learning

Table 7

Weights model vs. baseline.

Datasets	Model	Test RMSE	Test MAPE	Validation RMSE	Validation MAPE
P100_Conv	Baseline	2.962 ms	14.36%	3.455 ms	15.17%
	Weights model	2.894 ms	15.24%	3.091 ms	14.58%
V100_Conv	Baseline	1.722 ms	11.44%	1.547 ms	11.13%
	Weights model	1.660 ms	11.03%	1.850 ms	11.49%
K40_Conv	Baseline	10.136 ms	15.59%	10.361 ms	16.39%
	Weights model	9.051 ms	15.14%	11.675 ms	15.16%
P100_Dense	Baseline	0.033 ms	3.15%	0.032 ms	3.04%
	Weights model	0.032 ms	3.05%	0.032 ms	3.02%
V100_Dense	Baseline	0.051 ms	6.26%	0.046 ms	5.84%
	Weights model	0.046 ms	6.07%	0.047 ms	6.10%
K40_Dense	Baseline	0.157 ms	7.57%	0.179 ms	7.92%
	Weights model	0.159 ms	7.11%	0.150 ms	7.07%
All_Conv	Baseline	4.079 ms	11.44%	4.021 ms	11.16%
	Weights model	4.273 ms	10.40%	3.893 ms	10.39%
All_Dense	Baseline	0.077 ms	4.70%	0.074 ms	4.68%
	Weights model	0.077 ms	4.72%	0.076 ms	4.73%