Ask HN: Can the following algorithm (written in Elixir) have applications in ML?

In a neural network training, the goal is to optimize the parameters such that the result is closest to what we want them to be. For example, we might want a model which can predict a continuous variable (like the price of a house) given a set of input variables.

When we actually compute the values of the neural network, it amounts to a large number of matrix multiplications. In fact we do parallelize this on a GPU. The thing we are trying to optimize is to minimize the loss, but it has less to do with finding a short path through the neural network layers.