Blind Federated Learning without initial model

Journal of Big Data

Table 22 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%	69%	0.9722	0.9722	0.9630	0.9630
2	20%	78%	0.8889	0.9722	1.0000	1.0000
3	20%	86%	0.9444	0.9722	1.0000	1.0000
4	20%	78%	0.9444	0.9444	0.9355	0.9355
5	20%	69%	0.8889	0.9444	0.8571	0.9231
Avg.	–	–	0.9278	0.9611	0.9511	0.9643
1	13%	63%	0.3834	0.4233	0.8667	0.8846
2	30%	80%	0.3140	0.3719	0.8182	0.7857
3	21%	87%	0.4182	0.4303	0.9194	0.9020
4	11%	61%	0.3462	0.3718	0.8750	0.9286
5	24%	81%	0.3175	0.3794	0.7700	0.8434
Avg.	–	–	0.3559	0.3953	0.8498	0.8688
1	4%	75%	0.3023	0.3256	1.0000	1.0000
2	5%	68%	0.2319	0.3188	0.8000	1.0000
3	45%	77%	0.3191	0.3418	0.7479	0.7705
4	13%	51%	0.4681	0.4681	0.5769	0.4500
5	32%	85%	0.2766	0.2901	0.9189	0.9057
Avg.	–	–	0.3740	0.3970	0.7478	0.6386
1	1%	25%	0.6667	0.6667	0.7143	0.6667
2	5%	72%	0.2949	0.3462	0.8750	1.0000
3	45%	78%	0.2889	0.3306	0.7727	0.8415
4	14%	53%	0.5829	0.6203	0.9091	0.9615
5	35%	89%	0.2893	0.3092	0.9625	0.9710
Avg.	–	–	0.4245	0.4546	0.8467	0.8881