Federated Learning of User Verification Models Without Sharing Embeddings

We consider the problem of training User Verification (UV) models in\nfederated setting, where each user has access to the data of only one class and\nuser embeddings cannot be shared with the server or other users. To address\nthis problem, we propose Federated User Verification (FedUV), a framework in\nwhich users jointly learn a set of vectors and maximize the correlation of\ntheir instance embeddings with a secret linear combination of those vectors. We\nshow that choosing the linear combinations from the codewords of an\nerror-correcting code allows users to collaboratively train the model without\nrevealing their embedding vectors. We present the experimental results for user\nverification with voice, face, and handwriting data and show that FedUV is on\npar with existing approaches, while not sharing the embeddings with other users\nor the server.\n