Incredibly hard for a newcomer like me, yet extremely interesting. I'd like to read your scribbled pen-and-paper.
I am looking for the topic of my master's thesis regarding AGIs but I have received proposals regarding Operational Neural Networks and if I could find their possible application in this field, in order to subsequently obtain a possible post-master study. Unfortunately I don't think the two can coexist for an AGIs base applied to OpenCog (right?) (I would strongly like to bring something related to OpenCog as a thesis) So I'm looking for an alternative in your project, but I keep getting lost in the topics I see from posts, papers, conferences and codes ... they are all very interesting. The point is that I'm kindly asking for help in finding something to develop to be useful for this project. Someone who gives me a more or less defined goal that I could aspire to ... because, despite being able to understand the problems and possible solutions you propose, I cannot then translate the words into a lineup of steps to follow to develop and implement them in the current project. Thank you for the trouble with my request. Michele Thiella Il giorno mercoledì 11 novembre 2020 alle 06:34:26 UTC+1 Ben Goertzel ha scritto: > Wow... this is possibly the most elegant Haskell I have ever seen... > check out the section titled "Horticulture with Futumorphisms" ... > > https://blog.sumtypeofway.com/posts/recursion-schemes-part-4.html > > Extreme simplicity, built on top of beautiful abstraction > > I have worked out (but not yet typed in) how to define fold and unfold > (catamorphisms, anamorphisms, histomorphisms and futumorphisms) on > hypergraphs and metagraphs. > > (By extending the ideas in > > https://www.researchgate.net/publication/2423982_An_Initial-Algebra_Approach_to_Directed_Acyclic_Graphs > -- but note the constructors one needs for metagraphs are a bit > different than the ones he describes for dags, though it seems his > basic framework still works) > > I believe one can then express > > * activation spreading (ECAN), imposing probability distributions on > Atomspace (for sampling), and crossover + mutation, as metagraph > catamorphisms > > * logical inference chaining and program execution as metagraph > futumorphisms (anamorphisms with built-in control structures) > > * inference "direct evaluation" as a metagraph histomorphism (a > catamorphism that keeps track of its history) > > If this is right (and of course some gotcha might emerge in > transferring my scribbled pen-and-paper notes into LaTeX, which I have > no idea when I'll have time for...) then we have a fairly > straightforward mapping of our repertoire of OpenCog cognitive > algorithms as metagraph hylomorphisms. > > Since hylomorphisms can represent general recursion, it's not > surprising (actually obvious) that one can represent these cognitive > algorithms as metagraph hylomorphism somehow, but the point is that it > can be done simply and elegantly... > > What I'm trying to work toward here is a sort of abstract requirements > spec for what Atomese 2 needs to do to yield efficient implementations > of relevant cognitive algorithms. I think the crux is that we need > to be able to reasonably efficiently do these sorts of morphisms over > Atomspace, in instances where the atomic operations inside the > morphism are small and inexpensive (so that a substantial fraction of > the work is in executing the morphisms rather than doing costly > operations on each Atom). > > > > -- > Ben Goertzel, PhD > http://goertzel.org > > “Words exist because of meaning; once you've got the meaning you can > forget the words. How can we build an AGI who will forget words so I > can have a word with him?” -- Zhuangzhi++ > -- 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 view this discussion on the web visit https://groups.google.com/d/msgid/opencog/86fc56e7-15bf-4838-923a-0cb4331e7866n%40googlegroups.com.
