Cache-Aided Matrix Multiplication Retrieval
share
Coded caching is a promising technique to smooth out network traffic by storing part of the library content at the users' local caches. The seminal work on coded caching by Maddah-Ali and Niesen (MAN) showed the existence of a global caching gain, in addition to the known local caching gain in uncoded systems, when users aim to retrieve a single file from the library. This paper formulates a novel cache-aided matrix multiplication problem. Matrix multiplication is an essential building block for distributed computing and machine learning applications. Different from the original coded caching model, in the considered problem each cache-aided user requests the product of two matrices from the library which contains N matrices. A direct solution which is agnostic the structure of matrix multiplication, is to treat each of the N^2 possible products as an independent file and use the MAN coded caching scheme with file retrieval. In order to improve the structure-agnostic scheme, by leveraging the correlation among the elements in each matrix product, two structure-aware achievable schemes are proposed, which partition each library matrix by rows and columns respectively, and let a subset of users cache each sub-matrix. Some order optimality results of the structure-aware achievable schemes are derived in the paper.
READ FULL TEXT