What is the main characteristic of federated learning?

• A. Centralization of raw data for training
• B. Models learning across multiple devices without sharing data
• C. Only using local data for training
• D. Real-time learning from a single source

Answer: (B)

The main characteristic of federated learning is that it allows models to learn across multiple devices without the need to share raw data. This approach is particularly advantageous because it enhances privacy and security by keeping data localized on the device where it originates. Each device trains the model on its local dataset, and only the updated model parameters are shared with a central server, which then aggregates these updates to improve the global model. This method ensures that sensitive personal data does not leave the device, thus addressing many privacy concerns associated with traditional machine learning approaches.

The focus on decentralized learning means that while each device contributes to the model’s overall performance, the raw data remains isolated and is not centralised, which mitigates the risks associated with data breaches or misuse. This characteristic is especially relevant in contexts such as healthcare and personal mobile devices, where privacy is paramount.

In contrast, other choices present concepts that do not align with the principles of federated learning. Centralization of raw data, reliance solely on local data without the broader collaborative benefit, and real-time learning from a single source are all contrary to the collaborative, decentralized approach that defines federated learning.