Hybrid Transformer + GNN intrusion detection model. Forward pass: 1. Transformer encodes the temporal sequence → H 2. GNN encode

Wraps Krum / Multi-Krum with an additional pre-filter step: 1. Compute cosine similarity of each update to the current global mode

Append-only immutable ledger that records every client's reputation update. Tampering with a block invalidates the chain (all subsequent

Simulates a single IoT edge device participating in federated learning. Parameters ---------- client_id : Unique integer identi

Central coordinator for the FedDefender FL simulation. Holds the global model and orchestrates all training rounds.

Single message-passing layer. h_v^(l+1) = σ( W^(l) · MEAN({h_u : u ∈ N(v) ∪ {v}}) + b^(l) ) This corresponds to Eq. 3 in the paper.

Maintains per-client reputation scores using Eq. 9 and writes every update to the blockchain ledger. Usage ----- rm = Reputation

Wraps (X, y) arrays and optionally builds a graph adjacency matrix for the GNN stream.