Learning internal representations in an attractor neural network with analogue neurons