Federated Learning (FL) is a machine learning technique that allows for the training of algorithms across multiple decentralized devices or servers while keeping the data localized. This method addresses privacy concerns by not requiring data to be shared with a central server, thus protecting sensitive information. FL involves multiple clients (such as mobile devices or organizations) that possess local data samples and perform computations on this data. These local computations are then aggregated to update a global model, managed by a central server, which coordinates the learning process and updates the clients with the improved model123.

The architecture of FL can be categorized into three types based on the data distribution among the clients: horizontal federated learning, vertical federated learning, and federated transfer learning. Horizontal FL is used when datasets share the same feature space but different samples. Vertical FL applies when datasets share the same sample space but differ in feature space. Federated transfer learning is utilized to transfer knowledge from one domain to another, helping to improve model performance when there are insufficient labeled data13. This structured approach allows FL to be applied in diverse scenarios, including healthcare, finance, and mobile services, where data privacy is paramount23.

Federated learning relies on several essential functions to facilitate collaborative model training across decentralized devices. At its core, federated learning involves two primary components: local training and global aggregation1. Local training occurs on individual client devices, where each client updates the model using its own data. The global aggregation step combines these local updates to improve the overall model.

Key functions in federated learning include:

• Model initialization: The central server initializes the global model and distributes it to participating clients2.

• Local computation: Clients perform model updates using their local data, typically through techniques like stochastic gradient descent1.

• Secure aggregation: The server aggregates local updates while preserving privacy, often using cryptographic techniques3.

• Model distribution: The improved global model is sent back to clients for the next round of training4.

• Client selection: In large-scale systems, the server may select a subset of clients for each training round to optimize resource usage5.

These functions work together to enable privacy-preserving, distributed learning across diverse datasets, making federated learning applicable in various domains such as mobile devices, healthcare, and finance32.

Federated learning offers significant advantages in data privacy and model diversity, but also presents unique challenges. One of the primary benefits is enhanced data privacy, as sensitive information remains on local devices, eliminating the need for centralized data storage12. This approach not only protects user privacy but also helps organizations comply with stringent data protection regulations2. Additionally, federated learning leverages diverse datasets from multiple sources, potentially improving model accuracy and robustness3.

However, federated learning is not without its challenges. The decentralized nature of the system makes it vulnerable to single-point failures, which can disrupt the entire learning process1. Communication costs can be substantial, especially when dealing with large-scale networks or complex models1. Furthermore, the non-independent and identically distributed (non-IID) nature of data across different clients can lead to model bias and convergence issues4. Addressing these challenges requires sophisticated algorithms and protocols to ensure efficient, secure, and effective model training in federated environments.