Nil, I'm thinking about moving forward with creating an Atomspace-based version of MOSES, with the work perhaps being split between Alexey's team in St. Petersburg and Kasim and Yidne's team in Addis Ababa...
You are of course aware that Yidne has been working on porting Reduct to the URE ... which is in a way the "hard part" of porting MOSES to Atomspace As I mentioned before, the general design I'm thinking of is -- make a MOSES version where each MOSES deme is an Atomspace -- keep the rest of MOSES as-is ... i.e. the deme management, the feature selection, etc. etc. So basically the actual collection of program trees being grown inside a deme (extending an exemplar) would live in an Atomspace.... For a given MOSES "run", one could perhaps have one central Atomspace accumulating the top N programs from each of the peripheral Atomspaces, where each deme is a peripheral Atomspace.... Fitness evaluation should not be tricky here, basically a fitness function can be a GroundedSchemaNode Nil, my basic idea here (as we discussed once before) is that the growth of an exemplar program into various "derived" programs (as occurs within a MOSES deme) should be executable using a URE rule (or a small number of URE rules) .... I am thinking that your recent work on the URE Pattern Miner, should put you in a position where you can clearly articulate how to use the URE to implement the growth of a population (deme) of programs from an exemplar program... (Obviously the big intended payoff here is a return to the roots of MOSES, i.e. the use of probabilistic analysis to find patterns regarding which program-trees in a deme are successful. I would like to be able to use pattern mining and PLN to help with this, and this will be much easier to experiment with once the MOSES program trees are in the Atomspace.... But I'm not looking to implement/design the details of this aspect yet; first I want to just deal with the "mechanics" of getting a MOSES deme to be an Atomspace...) Anyway what I am hoping is that you can write a sort of spec for "how to use the URE to implement the growth of programs from the exemplar in a MOSES deme" .... if you can viably do this before we meet F2F in HK in March that would be ideal for spurring discussions and moving the idea to the next stage... (A note to others: It is clear that the mechanics of MOSES program execution will be slower in the (current version of the) Atomspace than in the current version of MOSES. However, the Atomspace brings with it other possibilities, including the caching of partially evaluated results, as well as easier probabilistic modeling as mentioned above.... And of course we can cook up various schemes for accelerating Atomspace program execution if we want to. Generally though I am thinking the value-add of Atomspace-MOSES will occur mostly for cases where fitness evaluation is highly expensive for reasons other than having a large number of simple operations carried out in the program tree, e.g. where the bulk of time that the program spends is in manipulation of Atoms or manipulation of external data, in which cases the slower mechanics of program tree evaluation in Atomspace as opposed to VertexTrees doesn't matter...) thanks Ben thanks Ben -- Ben Goertzel, PhD http://goertzel.org "In the province of the mind, what one believes to be true is true or becomes true, within certain limits to be found experientially and experimentally. These limits are further beliefs to be transcended. In the mind, there are no limits.... In the province of connected minds, what the network believes to be true, either is true or becomes true within certain limits to be found experientially and experimentally. These limits are further beliefs to be transcended. In the network's mind there are no limits." -- John Lilly -- 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 [email protected]. To post to this group, send email to [email protected]. 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/CACYTDBeYvN%2BGcgnMXhrQ-Bjo%3D0tDiAhVMJYm-dZTGHYJ6dUmww%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
