Название: What Is Federated Learning?
Автор: Emily Glanz, Nova Fallen
Издательство: O’Reilly Media, Inc.
Год: 2021-10-14
Язык: английский
Формат: epub
Размер: 10.2 MB

The use of Artificial Intelligence (AI) and Machine Learning (ML) technologies is growing rapidly. AI and ML are empowering many application domains, such as natural language processing (NLP), image classification, and recommendation systems. You’ve probably seen this yourself: smartphone keyboard autocorrect features, face unlock capabilities, and movie recommendation algorithms are all powered by Machine Learning. One of the key reasons for the success of ML technologies today is the accessibility and availability of massive amounts of data.

The data used to train a given model has traditionally been centralized and stored in a single location. However, there are many domains where it is difficult and/or undesirable for the training and evaluation data to be centrally collected. Centralization of data also raises privacy concerns. Data ownership issues have moved to the forefront of the public consciousness, which has brought new scrutiny to how data is collected, stored, and used for traditional machine learning.

What if organizations didn’t have to collect data and store it in a central location in order to train a model with Machine Learning? The key idea behind Federated Learning (FL) is that it is possible to bring model training to the location where the data was generated and lives, removing the requirement for centralized data collection. Federated Learning is a type of Machine Learning in which a set of clients collaboratively train a model on local data under the orchestration of a central server, without sharing raw data with each other or the server. Instead, the clients only share model updates focused on the Machine Learning task. This allows the power of data at the edge to be harnessed, while preserving the privacy of the individuals or groups that generated the data.

Federated Learning is a compelling and increasingly popular approach for training Machine Learning models on sensitive data in a privacy-preserving manner. As you’ll see in Chapter 2 of this report, it’s compatible with a number of other privacy technologies. These technologies can be used to guarantee that the model updates that clients transmit during the FL training process are encrypted until they are aggregated, and that it is not possible to recover client data from the final trained model. By combining Federated Learning with other privacy technologies, it is possible to provide strong privacy and anonymization guarantees regarding how individual data is processed. In this book, we will provide motivating examples for federated learning, examine the lifecycle of training a model with federated learning, and explore the unique challenges of training a model in a privacy-preserving manner.

Until recently, an organization would have had to collect and store data in a central location to train a model with Machine Learning. Now, Federated Learning offers an alternative. With this report, you'll learn how to train ML models without sharing sensitive data in the process. Google software engineers Emily Glanz and Nova Fallen introduce the motivation and technologies behind Federated Learning, providing the context you need to integrate it into your use cases.