In this talk, we will review some recent results on coded distributed matrix multiplications, where the goal is to perform matrix multiplication over a set of N worker nodes, so that the result can be recovered from any K workers. We describe systematic code constructions, where each node stores a fraction of 1/m of the two matrices involved, and obtains a recovery threshold of K=2m-1. An interference alignment perspective of the constructions is provided.

We also summarize some code constructions and interesting open questions that arise when the matrix alphabet is restricted to binary alphabet, and computationally efficient matrix multiplications algorithms are utilized.

The talk is based on several results obtained through collaborations with Mohammad Fahim and Farzin Haddadpour at Penn State, and Sanghamitra Dutta, Haewon Jeong and Pulkit Grover at Carnegie Mellon University.