Blind Federated Learning without initial model

Journal of Big Data

Table 28 Blended Blind FL results (Precision-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%	72%	0.8889	0.9444	0.8966	0.9000
2	20%	72%	0.8889	1.0000	0.9630	1.0000
3	20%	81%	0.9722	0.9722	1.0000	1.0000
4	20%	69%	1.0000	1.0000	1.0000	0.9688
5	20%	83%	0.9167	0.9722	0.9000	0.9643
Avg.	–	–	0.9333	0.9778	0.9519	0.9666
1	13%	63%	0.3007	0.3277	0.7692	0.8205
2	30%	80%	0.3651	0.3730	0.8696	0.8182
3	21%	87%	0.2708	0.3095	0.7045	0.7500
4	11%	61%	0.3618	0.3750	0.9524	1.0000
5	24%	81%	0.3115	0.3496	0.8429	0.9063
Avg.	–	–	0.3220	0.3470	0.8277	0.8590
1	4%	75%	0.2785	0.3038	0.6667	0.6000
2	5%	68%	0.3200	0.4000	0.7778	1.0000
3	45%	77%	0.3324	0.3709	0.9157	0.9740
4	13%	51%	0.4433	0.4588	0.7500	0.7500
5	32%	85%	0.3062	0.3159	0.9722	0.9773
Avg.	–	–	0.3361	0.3699	0.8165	0.8603
1	1%	25%	0.6875	0.7188	0.6667	0.5000
2	5%	72%	0.2692	0.2949	0.8182	0.7143
3	45%	78%	0.3045	0.3361	0.7232	0.7973
4	14%	53%	0.5000	0.5412	0.7419	0.7368
5	35%	89%	0.3145	0.3183	0.9271	0.8889
Avg.	–	–	0.4152	0.4418	0.7754	0.7275