🧪 Federated Learning Simulation

Federated learning trains one shared model across many devices without ever collecting their raw data — but real devices rarely hold identical data (non-IID), so a single global model is always a compromise. This simulator runs a real FedAvg loop across six clients with their own rotated decision boundaries: each round every client takes local logistic-regression gradient steps and the server averages the resulting weights into a new global model. A personalization slider then blends each client's local fine-tuned weights back in, so you can watch the accuracy/generalization trade-off play out live as points flip between correctly and incorrectly classified.