Structured Polynomial Codes for Distributed Matrix Multiplication over Correlated Sources
A coding scheme with matching information-theoretic bounds for computing bilinear functions (dot products and matrix products over finite fields) from two correlated distributed sources. Students learn how nonlinear source transformations combined with Korner-Marton structured linear coding can beat Slepian-Wolf compression, and how this yields structured polynomial codes that improve distributed matrix multiplication in the master-workers-receiver setting.
Structured Codes for Distributed Matrix Multiplication
Our work addresses the well-known open problem of distributed computing of bilinear functions of two correlated sources ${\bf A}$ and ${\bf B}$. In a setting with two nodes, with the first node havin…