The user embedding for a user u ( P u) and item embedding for item i ( Q i) are trained to predict the user’s rating for that item ( R ui). Left: A matrix factorization model with a user matrix P and items matrix Q.
Even for models without user-specific embeddings, having some parameters be completely local to user devices would reduce server-client communication and responsibly personalize those parameters to each user. Training a fully global federated model would involve sending user embedding updates to a central server, which could potentially reveal the preferences encoded in the embeddings.
Consider models with user-specific embeddings, such as matrix factorization models for recommender systems. However, in some settings privacy considerations may prohibit learning a fully global model. This heterogeneity has motivated algorithms that can personalize a global model for each user. For example, users of a mobile keyboard application may collaborate to train a suggestion model but have different preferences for the suggestions. Often this is done by learning a single global model for all users, even though the users may differ in their data distributions.
#Project magenta demo google software
Posted by Karan Singhal, Senior Software Engineer, Google Researchįederated learning enables users to train a model without sending raw data to a central server, thus avoiding the collection of privacy-sensitive data.
0 kommentar(er)
