[stack] Re: Evolutionary Programming

Did anybody else follow up on the good links which 
ementation 2 gave ?

> Jürgen Schmidhuber starts laying out the algorithms in: 
> http://www.idsia.ch/~juergen/mljssalevin/node5.html
> 
> His drastic claims seem plausible.
> 
> So far I haven't seen what language/s he proposes.

== Chris Glur

PS. I'm testing the hack: post via gmail without bogging
down in web-click-mechanism.

My 'proper' mailing-list gives auto-confirms, but perhaps
this one is moderated ?
--------
In: http://www.idsia.ch/~juergen/mljssalevin/node9.html
   he tells: ---------
              Implementation 2: Incremental Self-Improvement (IS)
..snip...
=  Overview.  We  will  use a simple, assembler-like programming language
   which   allows  for  writing  many  kinds  of  (learning)  algorithms.
-------- and in node10.html  he tells:-------
   $b_0$ :
          Add(  $w_1,w_2,w_3$  )  : $c_{w_3}\leftarrow c_{w_1} + c_{w_2}$
          (add  the  contents  of  work  cell $w_1$ and work cell $w_2$ ,
          write the result into work cell $w_3$ ).
   $b_1$ :
          Sub( $w_1,w_2,w_3$ ) : $c_{w_3}\leftarrow c_{w_1} - c_{w_2}$ .
   $b_2$ :
          Mul( $w_1,w_2,w_3$ ) : $c_{w_3}\leftarrow c_{w_1} * c_{w_2}$ .
   $b_3$ :
          Mov( $w_1, w_2$ ) : $c_{w_2}\leftarrow c_{w_1}$ .
   $b_4$ :
          JumpHome: IP $\leftarrow 0$ (jump back to 1st program cell).

   Instruction  probabilities / Current policy. For each program cell $i$
   there  is a variable probability distribution $P_i$ on $I$ .
----
That's apparently Latex format, where $b_1$ looks like
 "b <subscript 1>".   But you get the flavour ?

IMO this all looks promising.

BTW I think: c1 + c2 -> c3 is a better syntax.
Pop-11 uses it and I just can't see why "c3 <- c1 + c2", has
prevailed.

== Chris Glur.


ET.Do  CatGmailSET