Batch-Aggregate: Efficient Aggregation for Private Federated Learning in VANETs
Xia Feng, Haiyang Liu, Haowei Yang, Qingqing Xie, Liangmin Wang
Abstract
Federated learning (FL) in Vehicular Ad-hoc Networks (VANETs) enables vehicles to collaboratively train machine learning models by aggregating local gradients without revealing the training data. To ensure no gradient is revealed during aggregation, proposals are using a secret sharing-based strategy. A major bottleneck for applying these proposals in VANETs is the overhead of model aggregation across high-mobility vehicles. Particularly, the communication overhead grows exponentially due to the dynamic of VANETs. In the paper, we propose Batch-Aggregate, an efficient aggregation scheme for FL coping with high mobility and unstable connections of VANETs. By encoding the linear encryption into a short group signature, we combine authentication into aggregation protocol. When a registered vehicle trains its local model and sends the masked gradients to the nearby Road-side Unit (RSU), the RSU can independently check the gradients for validity and aggregate the parameters in a batch way. Thus, the computation time of the aggregator will be reduced to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mathcal {O}(n)$</tex-math></inline-formula> while the gradients can be aggregated in one communication round per training iteration. Moreover, our scheme provides privacy properties such as anonymity and unlinkability. The simulations show that the computation overhead of Batch-Aggregate grows linearly under the batch-enabled scheme, which reduces up to 50% over the existing schemes.