What is the term for a machine learning approach that trains a model across multiple decentralized devices while preserving data privacy?

The term that accurately describes a machine learning approach that allows training a model across multiple decentralized devices while maintaining data privacy is federated learning. This technique enables devices, such as smartphones or IoT devices, to collaboratively learn a shared prediction model while keeping all the training data local on the individual devices.

In federated learning, only the model updates are sent to a central server (not the raw data itself), which aggregates these updates to improve the global model. This approach significantly enhances data privacy, as personal data remains on the device and is not shared with the server or other devices.

In contrast, distributed learning typically involves splitting the data into chunks that can be processed in parallel, often requiring data to be pooled in a centralized location, which may compromise privacy. Collaborative filtering is a recommendation system technique used for predicting user preferences based on past interactions and does not focus on data decentralization or privacy concerns. Centralized learning relies on a single server to process all data, which is the opposite of what is intended in scenarios requiring data privacy. Thus, federated learning stands out as the correct answer in the context of preserving data privacy while enabling collaborative model training.