Distributed AI Cryptography & Consensus Lab
Enterprise Autonomous Agent Federated Secure Aggregation & Byzantine Consensus Simulator
Simulate and benchmark multi-agent federated learning resilience against gradient poisoning, sign flipping, and covert backdoors. Test cryptographic Secure Aggregation (SecAgg via Shamir Secret Sharing) and Byzantine-robust aggregation algorithms (Multi-Krum, Bulyan, Trimmed Mean) to ensure DPDP Act 2023 compliance.
Byzantine Aggregation Algorithm Resilience Matrix
- Standard FedAvg: 0% Byzantine resilience. Fails catastrophically with a single compromised node (f ≥ 1), causing global model divergence.
- Coordinate-wise Median: Resilient up to f < N/2 against independent coordinate noise shocks.
- Multi-Krum: Selects vectors minimizing Euclidean distance to nearest neighbors, resilient when 2f + 2 < N.
- Elesium Bulyan + SecAgg Architecture: Military-grade resilience (4f + 3 ≤ N) eliminating stealth collusive vectors while pairwise Diffie-Hellman blinding masks ensure 0.0% raw gradient leakage under DPDP Act Section 8(4).