We consider the problem of private distributed matrix multiplication under limited resources. Coded computation has been shown to be an effective solution in distributed matrix multiplication, both providing privacy against the workers and boosting the computation speed by efficiently mitigating stragglers. In this work, we propose the use of recently-introduced bivariate polynomial codes to further speed up private distributed matrix multiplication by exploiting the partial work done by the stragglers rather than completely ignoring them. We show that the proposed approach reduces the average computation time of private distributed matrix multiplication compared to its competitors in the literature while improving the upload communication cost and the workers' storage efficiency.
翻译:我们认为,在资源有限的情况下,私人分布式矩阵的乘法问题。编码计算已证明是分布式矩阵乘法的有效解决办法,既能提供对工人的隐私,又能通过有效减少排减速度来提高计算速度。在这项工作中,我们建议使用最近推出的双轨混合代码,通过利用排减器所做的部分工作而不是完全忽视它们来进一步加快私人分布式矩阵的乘法。我们表明,拟议办法减少了私人分布式矩阵乘法的平均计算时间,而文献中的竞争者则减少了私营分布式矩阵乘法的平均计算时间,同时提高了上传通信成本和工人储存效率。