Chapter 11: ByzSecAgg — Byzantine-Resilient Secure Aggregation

Learning Objectives

• Formalize the Byzantine threat model for federated aggregation: malicious users sending corrupted gradients
• Identify why plain Bonawitz secure aggregation fails against Byzantine adversaries
• Construct the CommIT-group ByzSecAgg protocol combining ramp secret sharing, coded outlier detection, and vector commitments
• Analyze the protocol's communication complexity $O(n \log n + B d)$ and its Byzantine tolerance $B$
• Compare with prior Byzantine-resilient aggregation schemes (Krum, Median, Trimmed Mean) and their $B$-tolerance
• Recognize the role of each primitive — ramp sharing for privacy efficiency, coding for outlier detection, commitments for integrity