Short paper: Secure Lightweight Computation for Federated N-Gram Language Model

We present a secure lightweight computation for federated data N-gram language models, which utilizes the Pailier encryption and 0-additive secret sharing. Our proposed scheme is simple and effective in terms of encryption and communication. Technically, each client masks the submitted data by the shared value of 0-addictive secret sharing before sending it to the server. The server needs to collect the data across the client network to aggregate the final data. Eventually, we can protect the individual data and support the server in generating the global model. We experiment our work on a public data set to illustrate the performance.