Re: READER and Model Cards

Owen
much of this is above my level as well, but given enough time and
attention, nothing we cannot work out
the mind is expandable

It could be very useful to have XML wrappers around Model Cards - for all
the good reasons that we know what xml does
so the stack would consiste of a ML model, a human readable schema (the
model card) and some xml to make
it all exchangeable and interoperable so that the web can handle this
information
This would bring us closer to a web standard for neurosymbolic integration?

Have looked around and not found much on xml for model cards, please point
me otherwise

Now, there are several ways model cards can be wrapped in xml, and using
stratml or a modified form
of stratml, could be a way to go about it.  adapt stratml elements to
explain what a model does, how it does it
who are the stakholders, owners, policies, etc etc -  ummmm,   potentially
very useful

would you want to base the ModelML onto stratml> could do
would you want to continue to  call it stratml? maybe not,

as one of my teachers says, if we want to cross the ocean, we need to let
go of the shore
If you can manage to do this Owen, then I think stratML would live on for
some generations to come

Happy to brainstorm


On Wed, Feb 28, 2024 at 5:05 PM Owen Ambur <owen.ambur@verizon.net> wrote:

> Paola, much of this is beyond my level of interest and expertise.
> However, I'm glad I took the time to review the reference to StratML on
> page 6 of the AIKR CG's August 22 report
> <https://www.w3.org/community/aikr/wiki/images/7/7b/AI_KR_FIRST_REPORT_PUBLISHED_VERSION.pdf>
> .
>
> I'll look forward to learning how the elements of an AI/KR "model card
> <https://www.taylorfrancis.com/chapters/edit/10.1201/9781003310785-9/towards-web-standard-neuro-symbolic-integration-knowledge-representation-using-model-cards-paola-di-maio>"
> relate to those specified in the StratML standard (ISO 17469-1) and perhaps
> also the Integrated Value Network (IVN <https://stratml.us/docs/IVN.xml>).
>
> Owen Ambur
> https://www.linkedin.com/in/owenambur/
>
>
> On Tuesday, February 27, 2024 at 10:48:38 PM EST, Paola Di Maio <
> paola.dimaio@gmail.com> wrote:
>
>
> Greetings, AI KR CG Dragons
>
> When we stop learning, we start dying.... such is the life of the scholar
>
> François Remy, Kris Demuynck, Thomas Demeester, BioLORD-2023: semantic
> textual representations fusing large language models and clinical knowledge
> graph insights, *Journal of the American Medical Informatics Association*,
> 2024;, ocae029, https://doi.org/10.1093/jamia/ocae029
>
>
> https://academic.oup.com/jamia/advance-article/doi/10.1093/jamia/ocae029/7614965?searchresult=1
>
> This article is state of the art  *Merci Francois is packed with stuff we
> should get our heads around
> I d like to invite interested members to collaborate on Hugging Face
> One awesome way to figure out what ML does
> is ot use a model card
>
> *READ, STUDY!!*
> https://huggingface.co/docs/hub/en/model-card-guidebook
>
> Some time ago I  proposed  model cards as a mechanism for  neurosymbolic
> integration, that ise use declarative/symbolic KR (the model card) to
> describe and annotate a subsymbolic AI (the ML learning model and
> associated processes)
> ttps://
> www.taylorfrancis.com/chapters/edit/10.1201/9781003310785-9/towards-web-standard-neuro-symbolic-integration-knowledge-representation-using-model-cards-paola-di-maio
>
> Apologies for  the paywall, I promise to publish a free version of this
> but essentially
> it proposes what is explained in the report which was published ahead of
> the chapter
>
> https://www.w3.org/community/aikr/wiki/images/7/7b/AI_KR_FIRST_REPORT_PUBLISHED_VERSION.pdf
>
> Oh my, have I been busy in these last few years
>
> I have started a community on hugging face for this CG and
> THERE IS WORK TO BE DONE
>
> https://huggingface.co/W3CAIKRCG
>
> Please join, hack in!!
>
> PDM
>