Table 4 Global model accuracy.

From: RAIM: three-stage stackelberg game for hierarchical federated learning with reputation-aware incentive mechanism

Mechanisms

RAIM

RAIM-RS

RAIM-NO

Unreliable end devices

10%

30%

50%

10%

30%

50%

10%

30%

50%

MNIST

0.9823

0.9701

0.9350

0.9284

0.7711

0.6899

0.9201

0.7527

0.6737

EMNIST

0.9797

0.9712

0.9315

0.9258

0.7722

0.6864

0.9176

0.7496

0.6630

FEMNIST

0.9833

0.9488

0.9029

0.8311

0.7004

0.6012

0.8190

0.6990

0.5801

CIFAR10

0.6427

0.6193

0.5712

0.5723

0.5416

0.3987

0.5601

0.5412

0.3902

SVHN

0.8254

0.8031

0.7949

0.7622

0.6953

0.5928

0.7414

0.6693

0.5582

CINIC10

0.6422

0.6128

0.5710

0.5718

0.5351

0.3985

0.5675

0.5261

0.3896

Back to article page