columnDNN: a slightly different approach to a generic ML architecture

This neural network left-multiplies vectors to achieve outputs, meaning that the calculated gradient during backpropagation is also a column vector.