Discussion in 'Article Discussion' started by bit-tech, 10 Sep 2018.
Something i don't understand with these claims of chips that mimic how the brain works is how.
If my understanding of the brain is correct don't neurons connect to each other via interneurons and those connections become stronger as we 'learn' things, if so how does fixed hardware go about strengthening particular pathways, how it works is set in stone after all, wouldn't it be the software (firmware) doing the changing?
The Wikipedia article on spiking neural networks is a pretty good introduction to the concept.
That's correct. Depending on how the chip is designed (anything from "just a big RAM bank that one or more processors scan over" to "a massive number of really really dumb networked cores") the 'connection' between 'neurons' is either entirely in software (determined by a set of addresses stored by that 'neuron' to indicate what others it is 'connected' to) or a set of physical connections to all (or a preset subset) of other neurons of which most remain unused. The latter is basically never implemented in hardware* as it means most of the silicon is wasted doing nothing, and can easily be replaced by a bus that does the same function but takes up less die area.
*The exception being a handful of 'wetware' chips that dump a bunch of harvested neurons on top of a bunch of electrodes and let them form connections in response to stimuli. But at that point it;s less a chop and more head cheese.
From the Wikipedia article:
I wouldn't be surprised to see that emboldened by me bit 'edited' in future..
Separate names with a comma.