On Securing Berrut Approximated Coded Computing Through Discrete Cosine Transforms

Abstract

Coded computing is a reliable and fault-tolerant mechanism for implementing large computing tasks over a distributed set of worker nodes. While a majority of coded computing frameworks address accurate computation of the target functions, they are restricted to computing multivariate polynomial functions. To generalize these computing platforms to non-polynomial target functions, Jahani-Nezhad and Maddah-Ali recently proposed Berrut Approximated Coded computing (BACC), which was proven fault-tolerant against stragglers albiet with tolerable approximation errors on the target functions. Despite these benefits, there is no formal study on the security of BACC against worker nodes which report erroneous computations. To fill this research gap, we use a coding-theoretic approach to propose Secure Berrut Approximated Coded Computing (SBACC), which is resilient to stragglers and also robust to the presence of such untrusted worker nodes. One of the highlights of SBACC is the new choice of evaluation points for distributed computation which makes the well-known Discrete Cosine Transform (DCT) codes amenable to error detection and correction. To validate the new choice of evaluation points, first, we derive bounds on the accuracy of SBACC in the absence of untrusted worker nodes. Subsequently, to handle the presence of untrusted worker nodes, we derive bounds on the accuracy of SBACC and show that interesting optimization problems can be formulated to study the trade-off between the error correcting capability of the DCT codes and the accuracy of the target computation.