Efficient Quantization Mean Estimation for Distributed Learning

The increasing size of data has created a pressing need for protection of communication and data privacy, spurring significant interest in quantization. This article proposes a novel scheme for variance reduced correlated quantization that is designed for data with bounded support and distributed mean estimation. Our method achieves a theoretical reduction in the mean square error for fixed and randomized designs compared to the correlated quantization method under different levels and dimensions scenarios. Several synthetic data experiments were conducted to illustrate the effectiveness of the approach and to provide a reliable approximation of the reduced mean square error based on the theory. The proposed method was also applied to real-world data in different learning tasks, which yielded promising results. Supplementary materials for this article are available online.