No, the data is never shared to the server nor to the other clients, the data always remains private during the model training. During inference the model is served in your local environment thus no data is being shared.

Since the model is trained locally where the data resides there is only sharing of model parameters with each other. Model parameters are independent variables (compared to real data) that are learnt & tuned during model training.

We use a privacy preserving machine learning technique known as Federated Learning - that enables local models to be trained at the client side and perform model aggregation at the server side. This enables multi-parties to collaborate without sacrificing data privacy.

It's near to impossible to reverse engineer your data from model parameters since these variables are independent learnable values. It is technically & economically infeasible for someone to reverse engineer the model parameters to its original data.