Federated Learning
What is Federated Learning?
Federated LearningA distributed machine-learning paradigm in which many clients collaboratively train a model under a central coordinator while keeping their raw data on-device.
Federated Learning (FL) trains a shared machine-learning model across many participants — phones, hospitals, banks — without centralizing the training data. Each client computes a local gradient or model update on its own data and uploads only the update to an aggregator, which combines them (often via FedAvg) into a new global model. Because raw updates can still leak information about training data, modern deployments layer in cryptographic protections: secure aggregation, differential privacy, and sometimes homomorphic encryption or MPC. Google introduced FL for Gboard keyboard suggestions in 2017, and it is now widely used for healthcare imaging, fraud detection across institutions, and on-device personalization of LLMs.
● Examples
- 01
Gboard learns next-word predictions from millions of phones without sending typed text to Google's servers.
- 02
Hospital consortia jointly train diagnostic models on patient data that never leaves each institution.
● Frequently asked questions
What is Federated Learning?
A distributed machine-learning paradigm in which many clients collaboratively train a model under a central coordinator while keeping their raw data on-device. It belongs to the Cryptography category of cybersecurity.
What does Federated Learning mean?
A distributed machine-learning paradigm in which many clients collaboratively train a model under a central coordinator while keeping their raw data on-device.
How does Federated Learning work?
Federated Learning (FL) trains a shared machine-learning model across many participants — phones, hospitals, banks — without centralizing the training data. Each client computes a local gradient or model update on its own data and uploads only the update to an aggregator, which combines them (often via FedAvg) into a new global model. Because raw updates can still leak information about training data, modern deployments layer in cryptographic protections: secure aggregation, differential privacy, and sometimes homomorphic encryption or MPC. Google introduced FL for Gboard keyboard suggestions in 2017, and it is now widely used for healthcare imaging, fraud detection across institutions, and on-device personalization of LLMs.
How do you defend against Federated Learning?
Defences for Federated Learning typically combine technical controls and operational practices, as detailed in the full definition above.
What are other names for Federated Learning?
Common alternative names include: FL, Federated machine learning.
● Related terms
- cryptography№ 987
Secure Multi-Party Computation (MPC)
A family of cryptographic protocols that lets several parties jointly compute a function over their private inputs while revealing nothing beyond the output.
- cryptography№ 481
Homomorphic Encryption
An encryption scheme that allows computations to be performed directly on ciphertexts, producing encrypted results that match the operations on the underlying plaintexts.
- cryptography№ 859
Private Set Intersection (PSI)
A cryptographic protocol that lets two or more parties compute the intersection of their private sets while learning nothing about the elements that are not in common.
- cryptography№ 1265
Zero-Knowledge Proof (ZKP)
A cryptographic protocol in which a prover convinces a verifier that a statement is true without revealing anything beyond the validity of the statement itself.
- cryptography№ 249
Cryptography
The science of securing information through mathematical techniques that provide confidentiality, integrity, authenticity, and non-repudiation in the presence of adversaries.
- cryptography№ 1152
Threshold Cryptography
A class of cryptographic schemes in which a secret key is split across n parties so that any t of them — but no smaller subset — can sign, decrypt, or perform any other key operation.