Federated transfer learning (Glossary)

Federated transfer learning is a term from the fields of artificial intelligence and big data and is primarily used in the development of smart, data-based applications. In traditional machine learning, huge amounts of data are collected and processed centrally. Federated transfer learning takes a different approach: the data remains where it is generated - for example on smartphones or in different companies - and is not sent to a central location. Instead, the artificial intelligences "learn" locally and only share the results or models with each other.

The concept of "transfer learning" means that an AI model learns from certain experiences or data sets and transfers this knowledge to new, similar tasks. Federated transfer learning combines both and ensures that several different sources benefit from each other without passing on sensitive data.

An illustrative example: several hospitals would like to jointly train an AI to recognise illnesses earlier. However, they cannot share their patient data for data protection reasons. Federated transfer learning enables each hospital to learn locally and only share the learning results. This way, everyone benefits from the exchange without revealing private data.