Lets consider Information reduction assume something is represented by one x and by one y value so to describe one object, we require two variables now we want to reduce this to only one variable you can do this by fitting a function you reduce two values #(x,y)# into one value #x# and some function #y(x)# less accurate, but more understanding Now lets do the same with a neuronal Network Here we have one input and one output so its basically just a way of encoding a function This more general function we can train and get a similar result, but with a more complex function here the complexity is given by the network architecture