Quantification of the Leakage in Federated Learning
This addresses privacy risks for users in federated learning systems, but it is incremental as it builds on prior work showing gradient-based leakage.
The paper tackles the problem of data leakage in federated learning by analyzing gradients in an approximated logistic regression model, showing that complete training data can be leaked when inputs are binary (0 or 1).
With the growing emphasis on users' privacy, federated learning has become more and more popular. Many architectures have been raised for a better security. Most architecture work on the assumption that data's gradient could not leak information. However, some work, recently, has shown such gradients may lead to leakage of the training data. In this paper, we discuss the leakage based on a federated approximated logistic regression model and show that such gradient's leakage could leak the complete training data if all elements of the inputs are either 0 or 1.