Re: [stack] Evolutionary Programming

chris glur <[email protected]> wrote:
> If it's to run as an auto-evolving machine, then decoding,
> analysing & reverse-engineering the results is not realistic.
> And not a goal.

You do have to decide the relative fitness of the results, so some
amount of decoding results is needed.

> At the same time you want to use the cost effectivness and
> speed of a PC.
> In fact why bother with joy ?
> Just go as close as managable [without crashes] to the
> native instruction set of the machine/PC.

There's a reason why this hasn't been done -- CPU instructions sets
and architectures aren't amenable to mathematical modeling.

> For those who know PCs down to the register level,
> "divide by zero" is a completely artifical human concept.
> In fact "divide" is not know at the register level.
> [ It might be interesting to 'evolve' a divide machine
> instruction sequence, from the basic add, subtr shifts and
> various logical-operations and test/branch instructions ?!]

Are you thinking of some specific chip? All PCs have a division
instruction in their CPU, and ways to handle division by zero. Intel
and PowerPC both. Some embedded processors don't (I like the SeaForth
S40C18, for obvious reasons to people in this newsgroup), but those
aren't PCs.

> Similarly the concept of "loop" is just a human construct,
> invented for manageability purposes.

No, it's a machine concept.

> Of course the finer the granualarity the closer to a
> continuous/analog-computer and the less 'problems'
> with chaotic discontinuity.

Now I do like the idea of an analog computer; but "fine granularity"
won't give you that. Technically, proteins/DNA are analog, but they
have ENORMOUS chaotic discontinuity. In fact, I suspect you MUST have
chaotic discontinuity in order to make effective computations;
otherwise you have to expend a lot of energy in order to get results.

> == Chris Glur.

-Wm