Nil,

>
> Deduction system can be understood very broadly, and may encompass
> inferences based on PPL models as well.
>
> PLN definitely draws, at least in principle, the relationship between
> deduction and data.
>
> ATM in practice it's a bit lacking though, for instance the link between
> the TV
>
> Implication <TV>
>   P
>   Q
>
> obtained from instances of P and Q is forgotten after the inference. This
> should be corrected. Meaning the inference rule
>
> D ;; <- instances of P and Q
> |-
> Implication <TV>
>   P
>   Q
>
> should be more something like
>
> LinkBetweenDataAndImplication <TV>
>   D ;; <- instances of P and Q
>   Implication
>     P
>     Q
> d ;; <- new instance pair of P and Q
> |-
> LinkBetweenDataAndImplication <TV_update>
>   Cons
>     d
>     D
>   Implication
>     P
>     Q
>
> It would also provide an incremental way to calculate the TV as opposed to
> batch processing every time.
>
> It's kinda scary, computationally wise, but it seems to do well most
> inference traces need to be recorded, not just conclusions. Yet another
> meta-learning black hole...


This might be ok when we are talking about small-dimensional tasks, but I
don't think this is a good idea for real-world problems...
BTW, one my colleague (Vitaly Khudobahshov) has had some ideas regarding
that meta-computations should/can be carried out on low-dimensional
instances of problems, and then derived specialized solvers to apply to
higher-dimensional instances... So, maybe, it's ok to have
LinkBetweenDataAndImplication
...

-- 
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/CABpRrhz8fGEjswHTkp3ZuhTKYQLsHSKFj2JCicOORq%2BNXwBSYQ%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to