Federated Learning

Federated learning is a machine learning approach in which AI models are trained locally on individual devices or servers—without centralizing the raw data—and only the training results are aggregated centrally to update the global model. Because personal and confidential data never needs to leave the device, it achieves both privacy protection and distributed processing. In this setup, models are trained individually on user smartphones or edge devices, and only the learning outcomes—such as weights or parameters—are sent to a central server. The server aggregates and integrates these results, updates the global model, and redistributes it to the devices. This means individual local data never leaves the device, yet the AI model is continuously improved. Key advantages of federated learning: • Can build AI models while keeping data confidential • Reduces data transmission volume and lowers communication costs • Enables AI adoption in highly regulated fields such as healthcare and finance • Allows continuous model improvement from a large number of users In practice, Google has adopted federated learning for text input (Gboard) on Android smartphones, improving prediction accuracy while reflecting each user's typing patterns and protecting privacy. However, challenges remain, including differences in device performance and connectivity, and security risks from malicious participants (e.g., poisoning attacks). As demand for "providing advanced AI while protecting personal data" continues to grow, federated learning is expected to become a foundational technology across many domains—healthcare, mobile apps, IoT, and beyond.