This could be really interesting. Just thinking quickly I could see using the JavaSpaces Primary(Master)-Worker being used. Where the Primary hands out work for each of the workers to work on their part of the problem with the primary being the integrator. As you mentioned getting a shared data model representing the knowledge would be the difficult part and solutions may be best done on a domain-specific basis.
Would something like the below work as a proof of concept? *Problem Description* Annotation of an image with multiple workers in which each worker is specialized in a specific type of "recognition" (e.g., Dogs, People, Cars) *Flow* 1) A Primary sends out the image data to each worker. 2) Each Worker analyzes the image for it's specific type of recognition. 3) Each Worker sends back the location and type of object it detects back to the Primary 4) The Primary then analyzes the information and interprets the image *Google Protocol-Buffer Portion* - The Workers broadcast what they are capable of using a neutral format - All data is sent in Google Protocol Buffer Format - Individual Workers can translate the data into whatever language they run - Workers send back data in a similar format to a Primary that can use in its language On Fri, Mar 12, 2021 at 3:30 PM Phillip Rhodes <motley.crue....@gmail.com> wrote: > > Hi Phil, > > Can you elaborate a little bit on how you use JavaSpaces in AI Blackboard > > system space? I think having some examples might help us better flesh out > > this idea. I am also just curious from a research perspective. :-) > > Only in a conceptual sense. I haven't written any code to this end > yet. Not knowing how familiar anyone reading this post will be with > the basic idea of Blackboard Systems (or even Tuple Spaces as far as > that goes) I'll share these two links first: > > https://en.wikipedia.org/wiki/Blackboard_system > https://en.wikipedia.org/wiki/Tuple_space > > So my interest is in using Java Spaces as the "shared space" where the > "experts" (eg, "agents" in another vernacular) share information. The > devil, of course, is in the details. One big "detail" is working out > what the representation of "knowledge" put into the blackboard should > be. Something really fine-grained like RDF style triples could be an > option, but there are all sorts of knowledge representation languages. > And that's assuming you want something symbolic. An even deeper > question gets into how you might integrate symbolic / sub-symbolic > agents in a model like this. Unfortunately I have more questions than > I do answers. But generally speaking, this kind of "stuff" is where a > lot of my interest lies. > > > Phil > -- Jeremy R. Easton-Marks "être fort pour être utile"
