Re: Using Frames in Machine Learning

Hello,

I think that would be only a fraction of approaches that could be used for KR.
However, AI would have to be an integrated solution of multiple
approaches - symbolic + sub-symbolic learning that sits basically as
an in-between engine of representation learning, reasoning, and
probabilistic learning methods.

An integrated AI solution would be something like clarion, 4caps,
act-r, aixi, calo, cojack, epic, ikon, impala, lida, nars, oscar,
soar, society of mind, subsumption etc. that take inspirations in
cognitive abstractions.

Thanks,

Adeel



On Thu, 28 Jul 2022 at 04:53, Paola Di Maio <paola.dimaio@gmail.com> wrote:
>
> I am not that old really, but I learned AI before the current
> generation. You may call me old school in that respect.
>
> In my continued effort to catch up with and get to the bottom of the current AI KR challenges, I get to share glimpses with this CG
>
> The direction being pursued at the moment, briefly discussed with Richard Lea, is a possible standard for neurosymbolic integration using a card approach
>
> I have been developing model cards and I am finalizing a chapter for Taylor and Francis (accepted) where I propose the novel approach, and a possible way forward  for standardization here if we have enough members to promote the CG to WG
>
> To advance this proposal it is necessary obviously that interested participants are at a minimum familiar with AI KR
> (artificial intelligence knowledge representation) core concepts, such as frames
>
> Two seminal reference papers I d like interested folks to read before we can discuss the current developments
>
> A Framework for Representing Knowledge -
>
> M Minsky ยท 1974
> AVAILABLE ONLINE
>
> AND
> Studies in Cognitive Science
> 1975, FRAME REPRESENTATIONS AND THE DECLARATIVE/PROCEDURAL CONTROVERSYTerryWinograd
> I can share my copy for those interested
>
> future discussions will rely on the understanding of
> these papers
>
> P

Received on Thursday, 28 July 2022 04:16:08 UTC