On 12/18/2016 06:38 AM, Ben Goertzel wrote:
I have posted my design idea  (discussed in a recent email thread) on
the wiki site here

http://wiki.opencog.org/w/Reimplementing_MOSES_in_the_Atomspace_Atop_the_Pattern_Miner

The Backward Chainer can also be framed that way. :-) Even more so directly than what you've sketched on the SampleLink page.

The BC as currently implemented evolves atomese programs, where each program encodes a specific forward chaining strategies (FCS for short). There is a difficulty though, there are no easy way to evaluate the fitness of a FCS, either it proves the target or it doesn't. Or let's say that the fitness landscape is extremely chaotic, some FCS may prove nothing at all, while a tiny variation of it may prove our target completely.

BUT this can be overcome by meta-learning, i.e. learning a measure of success for FCS that are half-way there. So in this framework meta-learning would be used to reshape the fitness function, from a crisp chaotic one to a smooth regular one.

Nil


so it doesn't get lost

ben


--
You received this message because you are subscribed to the Google Groups 
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to opencog+unsubscr...@googlegroups.com.
To post to this group, send email to opencog@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/opencog/5878DC30.5080303%40gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to