When computing the multiplication of two matrices, a common technique is to parallelize the computation, that is, to use multiple computers in other to perform "different parts" of the multiplication at the same time. Some of these computers may finish their computations significantly before the others. As a result, the execution time of the slowest computer becomes a limiting factor and this effect is aggravated when the number of computers is large. If we could interpret the slower computers as erasures in a word of a code, we would be able to recover the missing information from the faster computers, obtaining a method to mitigate this effect.

In this talk we explain this problem and its solutions employing certain classes of evaluation codes. This is joint work with Umberto Martínez-Peñas.