Blind Federated Learning without initial model

Journal of Big Data

Table 24 Blended Blind FL results (accuracy-based weights, slope 5, sigmoid)—Adult dataset

Agent	Size	% 1s	Accuracy	Accuracy	Precision	Precision
Agent	Size	% 1s	Pre-FL	Post-FL	Pre-FL	Post-FL
1	20%	75%	0.8889	0.9444	0.9565	0.9231
2	20%	94%	0.9444	1.0000	1.0000	1.0000
3	20%	78%	0.8333	0.8889	0.8800	0.8276
4	20%	78%	1.0000	1.0000	1.0000	0.9600
5	20%	78%	0.7500	1.0000	0.7188	1.0000
Avg.	–	–	0.8833	0.9667	0.9111	0.9421
1	13%	63%	0.3292	0.3727	0.8485	0.9091
2	30%	80%	0.3857	0.4000	0.9200	0.9091
3	21%	87%	0.3539	0.3896	0.8837	0.8929
4	11%	61%	0.3663	0.3779	0.9231	0.8636
5	24%	81%	0.3205	0.3413	0.9333	0.8594
Avg.	–	–	0.3511	0.3763	0.9017	0.8868
1	4%	75%	0.3171	0.3780	0.6667	0.7692
2	5%	68%	0.3553	0.3684	0.9231	1.0000
3	45%	77%	0.3980	0.4162	0.9242	0.9259
4	13%	51%	0.5106	0.5426	0.8667	0.6667
5	32%	85%	0.2659	0.2897	0.9206	0.9184
Avg.	–	–	0.3694	0.3990	0.8603	0.8560
1	1%	25%	0.5926	0.5926	0.5000	0.5000
2	5%	72%	0.2987	0.3377	0.6429	0.6667
3	45%	78%	0.3205	0.3438	0.7900	0.8163
4	14%	53%	0.5213	0.5403	0.9565	0.9167
5	35%	89%	0.2538	0.2885	0.9000	0.9706
Avg.	–	–	0.3974	0.4206	0.7579	0.7740