Created
August 25, 2015 23:13
-
-
Save PtrMan/f948bc46c44548a7cd67 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
<sseehh_> on the opencog mailnig list ben mentioned wanting to make a grid-like version of atomspace for processing image data like destin was supposed to be used for | |
<sseehh_> i think something similar could apply to nars | |
<sseehh_> a grid of inheritance terms | |
<sseehh_> and apply a regular inference algorithm across them uniformly | |
<sseehh_> below nal4 or something | |
<sseehh_> like today with tony's example, i confined it to level 1 | |
<sseehh_> only inheritance statements | |
<SquareOfTwo__> so they gave up destin :D | |
<sseehh_> no idea but itamar wanted to commercialize that shit in the worst way | |
<SquareOfTwo__> thats bad | |
<sseehh_> probably sold it to the military and moved to another galaxy | |
<SquareOfTwo__> because some ideas of destin were valid for perception imho | |
<sseehh_> http://binatix.com/ | |
<sseehh_> copyrig 2012 nothing else | |
<sseehh_> i bet he's offworld and that email bounces | |
<SquareOfTwo__> hm for high level perception you need some form of NAL-8 for motions | |
<SquareOfTwo__> or a emulation of NAL-8 | |
<sseehh_> yeah but for the lowest level of vision or other percept processing, lesser nal levels would preprocess it | |
<SquareOfTwo__> ...some AGI's make it into the intergalctic space, others dont | |
<SquareOfTwo__> agreed | |
<sseehh_> mini-concepts | |
<sseehh_> we didnt yet explore my idea with the vector truth values | |
<SquareOfTwo__> on high level foundalis like concept networks still make sense i think | |
<sseehh_> as a way of encoding sensory data | |
<sseehh_> without destroying it | |
<SquareOfTwo__> no idea how, just try it :) | |
<sseehh_> %<0.9, 0.2, 0.3>;0.75% | |
<sseehh_> <pixel1 --> eye>. :|: %<r=0.945, g=0.255, b=0.154;0.95% | |
<sseehh_> <pixel1 --> eye>. :|: %<r=0.945, g=0.255, b=0.154>;0.95% | |
<SquareOfTwo__> ah | |
<sseehh_> <pixel2 --> eye>. :|: %<r=0.945, g=0.255, b=0.154>;0.50% | |
<sseehh_> <pixel2 --> eye>. :|: %<r=0.945, g=0.155, b=0.154>;0.50% | |
<sseehh_> etc | |
<SquareOfTwo__> but aren't these properties? | |
<sseehh_> these wouldnt mix ordinarily with scalar truths | |
<sseehh_> it is replacing the frequency value, which has always been assumed to be a scalar number betwee 0..1.0 | |
<sseehh_> with a generic type | |
<sseehh_> and i did this with MetaTruth.java | |
<sseehh_> but its not subclassed any other way yet | |
<sseehh_> confidence remains the same | |
<SquareOfTwo__> hm | |
<sseehh_> these values could be decomposed to the scalar form, | |
<sseehh_> but it would involve additional terms | |
<sseehh_> in the meantime, the actual data can be stored this way | |
<SquareOfTwo__> colors make only sense in context to other pixels | |
<sseehh_> and if reasoning chooses to decompose it, then it can be | |
<sseehh_> <<pixel2 --> eye> --> [red]>. :|: %0.954;0.50% | |
<sseehh_> but up to that point it doesnt need decomposed and we don need to flood the input system with such data | |
<sseehh_> it can be packed into its original vector for | |
<sseehh_> form | |
<SquareOfTwo__> but then you need a triple | |
<SquareOfTwo__> like | |
<sseehh_> thats what i meant it is not gaining anything computationally | |
<sseehh_> you could use a product too | |
<SquareOfTwo__> <pixel2 --> eye>. :|: %<r=0.945, g=0.155, b=0.154>:0.3;0.50% | |
<sseehh_> what is the 0.3 there | |
<SquareOfTwo__> certainty of course | |
<sseehh_> then what is the 0.50 | |
<SquareOfTwo__> freq | |
<sseehh_> no what im saying is that the frequency is the vector | |
<sseehh_> in this alternate form | |
<sseehh_> that the frequency is a generic type | |
<sseehh_> except we had always assumed it was 0..1.0 scalar | |
<SquareOfTwo__> that does make even less sense to me | |
<SquareOfTwo__> so there is no frequency? | |
<sseehh_> no, the frequency is a generic type | |
<sseehh_> it doesnt need to be a scalar number value between 0..1.0 | |
<sseehh_> it could be a strig | |
<sseehh_> string | |
<sseehh_> a vector | |
<sseehh_> a bitvector | |
<sseehh_> whatever | |
<SquareOfTwo__> yes of course | |
<sseehh_> but that only in the context of it bveign a scalar between 0.1.0 does the NAL inference rules apply | |
<sseehh_> in these alternate forms, we could define logic rules. but we dont need to | |
<SquareOfTwo__> but a rgb value can'T be translated to a frequency | |
<sseehh_> i'm saying it is | |
<sseehh_> but its not one that we have logic rules for | |
<sseehh_> and we dont need them .. this form only is a transofrmation | |
<sseehh_> a way of encoding the input | |
<SquareOfTwo__> frequency does tell the system how many cases it has objservered where something is the case | |
<SquareOfTwo__> like 0001001 | |
<sseehh_> before it gets decomposed to the natural 0...1.0 freq type | |
<SquareOfTwo__> the freq for 1 is 2/8 | |
<sseehh_> yes, in that logic system it does | |
<sseehh_> but so does confidence | |
<sseehh_> for example | |
<sseehh_> if i observe 0.75;0.50 twice | |
<sseehh_> it is not the 0.75 value that will be modified | |
<sseehh_> it is the 0.50 confidence that will | |
<sseehh_> 0.75 remains the same | |
<sseehh_> likewise if i observe red twice with 0.50 | |
<sseehh_> it is still red | |
<sseehh_> but higher confidence | |
<sseehh_> red is a frequency on the color spectrumj | |
<sseehh_> regardless if it's decomposed to RGB or a light wavelength | |
<SquareOfTwo__> sure it are scalars | |
<SquareOfTwo__> but they make sense just together | |
<sseehh_> NAL is designed to only work with the scalar values, i realize this | |
<sseehh_> what im saying is an implementation-specific modifciation | |
<SquareOfTwo__> i've nothing against comming up with new scalar logics | |
<sseehh_> its sort of like complex numbers | |
<SquareOfTwo__> yes i understand this | |
<SquareOfTwo__> and interpreting sensory data differently makes much sense | |
<SquareOfTwo__> like you could cluster colors by time and context | |
<sseehh_> it is just a way of representing this sensory data in a temporary state | |
<sseehh_> losslessly | |
<sseehh_> before the system decides to decompose it for actual inference | |
<sseehh_> which it may never do | |
<SquareOfTwo__> that makes a lot of sense | |
<sseehh_> but if it never does, at leats it wont waste the inference's processing | |
<SquareOfTwo__> but,...its not nars anymore :D | |
<sseehh_> and if it does, it will have the complete original data | |
<sseehh_> at any point in the future with which it decides to operate on it | |
<sseehh_> which is not likely if it was processed up front and then AIKR'd to hell | |
<sseehh_> thousand cycles later who knows how many bits of precision are lost | |
<sseehh_> no its not nars | |
<sseehh_> and i dont think i brought this up on the list and mentioned to pei |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment