A lot of people have trouble understanding how neural networks store and retrieve information. Still I believe it's fairly simple to explain (which doesn't mean they are equally easy to decipher).I think it's comparable to an audio fragment. An audio fragment exists of one or more sinuses, an audio wave of fixed amplitude and wavelength. The problem is taking the fragment apart, reducing it back to the individual waves.I see the information in a neural network the same way. It's stored in spaghetti kind of way. The different activation configurations of a network represent different pieces of information. And just like audio, the complexity of bigger networks is so big it becomes impossible to reduce a configuration to it's original elements.I don't think it's really impossible to dissect such a configuration, but it's just as difficult as reducing a piece of audio to it's individual sound waves.Maybe I should explain what a neural network is. But you can just as easily
google for it yourself. Or check
wikipedia.