Owen Nieuwenhuyse wrote:

> > If I'd ask you to prove that the result in that chart for 2*5 is 10,

> > would you go testing every single one of the 100 enteries if it matches

> > 10 or will you look into colmun 5, row 2 and see if that one fits the

> > value of 10?

>

> You didn't say it was that kind of table!

> It sounds a bit like a mixed visual metaphor, though. Do you have

> a practical example you could show me?



It's a multidimensional map of pixels. The example I showed is basicaly

2D, but is a 3D meaning map (one dimension is the one that sets

diffirent enteries appart; it is basic to all Meaning Maps).



It was once used toghather with a computer cluster, to combine weather

reports from diffirent parts of america. As far as I know about that,

the programers used a very leaky form of a Meaning Map, because it was

only displayed, not used to calculate. The number of dimmensions with

them was 25.



> Most of the cells would be empty??



Yes. Well depedning on how much data is in it.



> You have to make some compromises. Some "sweeping" is unavoidable.

> FATs are mainly "bad" on disk media due to physical latency.



No, the reason is diffirent: FATs reffer to one specific thing about the

content, namely: the filename. If I am, for example, looking for stuff

related to Windows 95 bugs, I still have to scan all the file contens.



Eventualy you figure out that the FAT would need to be a duplicate of

the contens to have all the information scanable, which is utterly

non-functional to the idea of a FAT.



> > No need for a human sitting next to the database, pushing his brains out

> > through his nose and ears, thinking how to make an optimal and bug-free

> > categorization for one particular database for one particular device.

>

> You can automate that, depending on object class regularity.

> This controls at what class-level you store your meta-data.

> 

> You can also create programs to do re-optimisation in the background.

> You will need some time off-line if you want to do a major reshuffle.

> Humans do this with sleep.



Off time, balh, sucks... Meaning Maps do it without any time consuming

off time.



In either case, your method would be CPU-cycle consuming and both CPU

fequency and time mean money, so it's money wasing.



> > Note that Meaning Map pixels do not magicaly hold information, they just

> > exist or they don't. The information they represent is all in their

> > location on the Meaning Map.

>

> -A more practical and complete example, please!



A pixel on the crossing between a value of 5 from connector A and a

value of 7 from connector B exists. This means that if we're getting the

value of 5 as the input from sensor A, we must output 7 tough the

connector B. If there are more pixels existing in the line of 5 from

connector A, any of the coresponding outputs is correct or appropriate.



> From my brief discussions/monologues on visual object recognition,

> it is apparent that there are a wild variety of data structures, methods,

> and record sizes required.



Indeed, you will notice that a **MEAINING** Map is for storing

**MEANING**. 



> > You mean you've got a better definition of the meaning of an apple to a

> > device with 4 senses?

>

> I could probably write a book about what I know about apples, and what

> I could determine from observing one particular apple.

> 

> Your example doesn't address these aspects.



If you couldn't sense anything beyond sweetness, sourness, colour and

hardness, your output would be noting more and noting more than what an

equivalent Meaning Map could do.



Why don't you give me an example 1/4 a page of the book you could write

about an apple and let me put that into a meaning map format without

loosing any information at all?



--



Don't feel bad about asking/telling me anything, I will always gladly

reply.



Digging for info? Try AI Meta Search:

Http://WWW.AIMetaSearch.Com



GTSC4 -- If nobody else wants to do it, why shouldn't we?(TM)

Meson & GTSC4 are now merged:

Http://WWW.MesonAI.Com

