Interpreting Neural Min-Sum Decoders