Figure 8.4
A diagram showing client devices feeding federated training into a central server, leading to model development and deployment, with callout boxes asking what private information an actor might learn at each stage.
Threat models for an end-to-end federated learning system. The goal of the system is to release some models and metrics to the model engineer, and eventually deploy a model to production. Thus, anonymous aggregation is essential for these released outputs of the computation. Data minimization approaches can address potential threats to the device, network, and server, e.g. improving security and minimizing the retention of data and intermediate results.

or Create an Account

Close Modal
Close Modal