Editing Federated Learning (section)

== <span style="color: #FFFFFF;">Remembering</span> ==
* '''Federated learning''' — A distributed ML approach where model training happens locally on client devices, and only model updates (gradients or weights) are shared, never raw data.
* '''Client''' — A participating device or institution that holds local data and trains a local model. Examples: smartphones, hospitals, banks.
* '''Server''' — The central coordinator that aggregates client updates, computes a new global model, and distributes it back to clients.
* '''Round''' — One communication cycle: server sends model → clients train locally → clients send updates → server aggregates.
* '''FedAvg (Federated Averaging)''' — The foundational FL algorithm by McMahan et al. (2017); aggregates client models by weighted averaging of parameters.
* '''Local epochs''' — The number of training epochs each client performs on local data before sending updates to the server.
* '''Non-IID data (Non-Independent and Identically Distributed)''' — A key challenge in FL: each client's data reflects its own distribution, which may differ substantially from other clients.
* '''Communication efficiency''' — Minimizing the amount of data transferred between clients and server, a key FL challenge.
* '''Gradient compression''' — Techniques to reduce the size of gradients transmitted (sparsification, quantization).
* '''Differential privacy (DP)''' — A mathematical privacy guarantee that limits what can be learned about any individual's data from shared model updates.
* '''Secure aggregation''' — Cryptographic protocols ensuring the server can compute the sum of client updates without seeing any individual update.
* '''Model poisoning''' — An attack where malicious clients submit corrupted updates to degrade or manipulate the global model.
* '''Byzantine fault tolerance''' — The ability of an FL system to produce correct results even when some participants are malicious or faulty.
* '''Cross-device FL''' — Federated learning across many mobile devices (millions of clients, heterogeneous, unreliable).
* '''Cross-silo FL''' — Federated learning across a small number of organizations (hospitals, banks), each with large datasets.
</div>

<div style="background-color: #006400; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">