Efficient Interactive Proofs for Linear Algebra

Motivated by the growth in outsourced data analysis, we describe methods for verifying basic linear algebra operations performed by a cloud service without having to recalculate the entire result. We provide novel protocols in the streaming setting for inner product, matrix multiplication and vector-matrix-vector multiplication where the number of rounds of interaction can be adjusted to tradeoff space, communication, and duration of the protocol. Previous work suggests that the costs of these interactive protocols are optimized by choosing O(log n) rounds. However, we argue that we can reduce the number of rounds without incurring a significant time penalty by considering the total end-to-end time, so fewer rounds and larger messages are preferable. We confirm this claim with an experimental study that shows that a constant number of rounds gives the fastest protocol.