FedBDQB: Communication Efficient Federated Learning via Bidirectional Dynamic Quantization and Bitmap

In this paper, we propose a novel framework for communication efficient federated learning, FedBDQB, by employing bidirectional dynamic quantization for model updates and bitmap-based lossless compression, and provide its descending analysis and convergence proof. The performance of FedBDQB is evaluated and compared against the classic FedAvg and a state-of-the-art method FedDQ across various experimental scenarios, including two widely-adopted datasets MNIST and CIFAR-10, as well as both IID and non-IID data partitioning. Experimental results demonstrate that our proposed FedBDQB can significantly reduce the data communication volume, achieving compression rates of up to 24.21 and 12.87 times over FedAvg and FedDQ, respectively. The substantial improvement in communication efficiency is attained with a negligible decrease in model accuracy (with the maximum error remaining below 1.5% across all experiments) and an acceptable increase in computational time (less than two times over FedAvg and FedDQ across all experiments).