Cortex supercharged by a neural network ...

'fraid not. Worked at Cybernetics dept. in Reading University years ago where we had a project which used a simple RAM based discriminator to quickly recognise and knock badly finished Brylcreem tubs off a production line . Also dabbled with neural networks in adaptive control context but not much beyond. My lasting impression is they are by no means some kind of panacea but certainly great in some pattern matching or even extrapolation applications. Could use in an IDRATEK context for example to implement visual person recognition or perhaps even 'unusual' trends in sensory data for some purpose. So I guess a case of identifying the purpose first (and then finding money to pay for R&D )

It looks to me like this particular device is still based on 'standard' processing technology, implementing more visual processing function units in parallel at hardware level and neural network type algorithms on top of all this, but it is not literally a neural network type of physical structure. Not that that is a critcism - just noting really

interesting ... would be good to use NN in some sort of watching role, to flag up when unusual things occur, and to optimise how some things are done ...

datasheet here :







a while back, worked with people who used NN to great effect ... key things needed, though, seemed to be lots & lots of data, from which to learn / tune, some sort of ability to quickly zap rouge data points / erroneous readings, experience & a knack in choosing optimum network topologies & weightings etc, and suitable situations ...