Table 11 Comparison of privacy protection metrics between non-federated and federated learning models.

From: Federated learning with differential privacy for breast cancer diagnosis enabling secure data sharing and model integrity

Metric

Non-federated learning model

Federated learning model (With DP)

Privacy Budget (ε)

No DP (Unprotected data exposure)

1.9 (Good privacy protection)

Gradient Sensitivity (L2 Norm)

No Clipping (Unrestricted gradients)

1.8 (Gradient clipping applied)

Membership inference attack risk

0.75 (High privacy risk)

0.52 (Better privacy)

Back to article page