Structure of the Optical Nerves ??

Eugene Leitl ui22204 at
Fri Sep 8 11:14:52 EST 1995

On 7 Sep 1995, Jan Vorbrueggen wrote:

> In article <42ldqb$n7b at> rising at (Hawley K. Rising III) writes:
>    >The problem is: if the retina's gone, you have to mimick the
>    >its function with a circuit/algorithm (I totally ignore interacing
>    >difficulties here). The information compression factor is 126:1, a lot 
>    >of processing horsepower (well beyond any current/near future 
>    >supercomputer according to Moravec. And he's too optimistic, imo). 
>    Why is this information compression, and why is it beyond a supercomputer?
>    I've seen no information which indicates that it goes 126:1 *retaining* all
>    the information, only that its 126:1.

It _is_ information compression since the bit rate goes down. 
I have not implied the compression to be lossless. Lossy compression
does not indicate the necessary amount of computation to be smaller.

I thought this was obvious. 

> Right. And one would try to do a lot of the imitiation of retinal
> preprocessing in analog hardware - no supercomputer required. Density is

Unfortunately, VLSI suffers from low integration density, low precision
and very low connectivity as compared to the mammal retina. Nature is still
_very_ superior here.

> another question. One could get approximately the same number of receptors on
> the same area, and sheer size is also almost there (you can buy 2k by 2k CCDs,
> that's only a factor of 30 below biology...). It would be a little difficult,

The sheer number of receptors is not the problem, agreed. However, the retina
is a very powerful computer, too. A CCD is not. (I am aware of recent analog 
silicon retinas from Nippon and elsewhere. Obviously what they do is a very
small beer in respect to the retina circuit).

> I guess, to imitate the different spectral sensitivities of the different
> receptor types, I gues, if you mix them on the same chip. And then you have
> the problem of producing a chip in the form of a speherical shell...

To coat each receptor with an according pigment layer is not the big problem.
And the chip may well remain flat if e.g. holographic optic is used.

-- Eugene
> 	Jan

More information about the Neur-sci mailing list