That is very true Stefan. Detector NN have layers of where simple features are generally in the first layers of the NN. And complex feature are detector are in the later part. There is bit of play in applying the back propagation algorithm during training so that complex detection and simple detection can be spread out over the many layers. IF the fist layer has a lot of neuron then it there is a greater chance the complex detection can happen tin the fist layer.
This can cause a trouble shooting nightmare. So if detection could be localized then NN would lose there "black box" reputation. The brain uses unsupervised detector NN. It also uses spikes and sine pules or waves. In my NN AGI model, Spike are to transmit information. Like that of a data on data bus, threw out the brain and optic nerve. Spikes pulses are lengthened into wave or a sine wave for mix together, fft. Mixing constructively and de destructively and with squashing function. This is information processing like that of artificial neural networks. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T4aa81fd9912dbd39-M83ca7c83a65edcb441ccd09e Delivery options: https://agi.topicbox.com/groups/agi/subscription
