|
Tuesday, 16 May 2017
quote [ Scientists dream of recreating mental images through brain scans, but current techniques produce results that are... fuzzy, to put it mildly. A trio of Chinese researchers might just solve that. They've developed neural network algorithms that do a much better job of reproducing images taken from functional MRI scans. The team trains its network to recreate images by feeding it the visual cortex scans of someone looking at a picture and asking the network to recreate the original image based on that data. ]
It's like a Convolutional Neural Neural Network. :D
|
midden said @ 6:58pm GMT on 16th May
That would surprise me, but would be awfully cool. It seems to me that everyone's visual cortex was trained and the neurons pruned in a unique sequence of sensory stimulations and feedback cues over many years. For instance, I wouldn't be at all surprised if someone who's native writing style was kanji had a vary different encoding of the shapes of the latin alphabet. Not that it would use a different principle of encoding, but that the particulars of the spacial neuron placements and connections could be very different within the visual cortex. On the other hand, maybe that's what makes training a digital neural network to identify these patterns so powerful; the fact that they recognize the forms based on the principles, regardless of the particular instantiation of those principles.
midden said @ 6:59pm GMT on 16th May
That would surprise me, but would be awfully cool. It seems to me that everyone's visual cortex was trained and the neurons pruned in a unique sequence of sensory stimulations and feedback cues over many years. For instance, I wouldn't be at all surprised if someone who's native writing style was kanji had a vary different encoding of the shapes of the latin alphabet, even if fluent in it. Not that it would use a different principle of encoding, but that the particulars of the spacial neuron placements and connections could be very different within the visual cortex. On the other hand, maybe that's what makes training a digital neural network to identify these patterns so powerful; the fact that they recognize the forms based on the principles, regardless of the particular instantiation of those principles.
<-- Entry / Current Comment
midden said @ 6:58pm GMT on 16th May [Score:1 Insightful]
That would surprise me, but would be awfully cool. It seems to me that everyone's visual cortex was trained and the neurons pruned in a unique sequence of sensory stimulations and feedback cues over many years. For instance, I wouldn't be at all surprised if someone who's native writing style was kanji had a vary different encoding of the shapes of the latin alphabet, even if fluent in it. Not that it would use a different principle of encoding, but that the particulars of the spacial neuron placements and connections could be very different within the visual cortex. On the other hand, maybe that's what makes training a digital neural network to identify these patterns so powerful; the fact that they recognize the forms based on the principles, regardless of the particular instantiation of those principles.
