W3C home > Mailing lists > Public > public-aikr@w3.org > May 2020

Re: Meeting Report

From: Paola Di Maio <paola.dimaio@gmail.com>
Date: Thu, 14 May 2020 19:36:16 +0800
Message-ID: <CAMXe=Sr6npk_=RXn613SBHEEsX20QCud9w2QTvnbQKHoi6-PGw@mail.gmail.com>
To: Owen Ambur <Owen.Ambur@verizon.net>
Cc: W3C AIKR CG <public-aikr@w3.org>
It is under Owen and Chris's leadership that we are making some progress
towards
adopting stratml for the AIKRCG 'strategy'

In sum. what are we doing/planning to do as a group is going to be
documented in the plan. and although we are still working  things out, as
we do have moments of brilliance and outbursts of productivity we can put
them down in this stratml plan on the stratnav app so that they ll be a
record of that. should be useful. I apologize again for being very tired
but 9 pm is very late for me. especially when I have had a full day incl
other calls etc-

a few notes below


the plan being developed here
> <https://www.stratnavapp.com/StratML/Part1/413d648b-bd36-418d-af74-e15b0cd8281d/Styled>.
> if anyone is inspired to chip in pls ask editing pass to Chris on this list
>


> With reference to our Frameworks goal
> <https://www.stratnavapp.com/StratML/Part1/413d648b-bd36-418d-af74-e15b0cd8281d/Styled#Goal_f1a62bb5-9910-4052-946a-344c0e22272f>,
> I will endeavor to render in StratML Part 2 format any frameworks that may
> be discovered and available on the Web.  Please apprise me of any of which
> you are aware.
>
To clarify -  Jorge aske whether we are using any framework of reference
for our work. which loosely attempts  to study explainability for machine
learning. That particular goal for our CG may need to be refined a little -
I dont think a frameworks exists as such (strategies, methods) but there is
interesting work being done, which I dont think is a framework yet. rather
a compilation of possible techniques. the effectiveness of which may need
to be evaluated in the field. So to answer the question, methods to address
explainability of ML exist but
a) I dont think are frameworks/strategies - this may be our goal? to gather
what is in the field and make a framework?
b) evaluation criteria for the effectiveness of these methods may not yet
be studied, again
could this be our work? I am doing some research in this direction but not
yet conclusive
 I volunteered to take up this task and shall soon update the plan with
some links but  I am putting together a presentation-
 anyone want to contribute?

The caveat is that  statistical pronability and non parametric methods in
ML are  unpredictable by definition
https://machinelearningmastery.com/uncertainty-in-machine-learning/
http://mlg.eng.cam.ac.uk/zoubin/talks/mit12csail.pdf

(this is not my field at all, does anyone care to expand?)
so I am not sure how to address this unpredictability other than with the
question

Can we use known symbolilc KR to  explain ML?

In the meantime, this Google site-specific query
> <https://www.google.com/search?ie=UTF-8&oe=UTF-8&q=AI+framework&btnG=Google+Search&domains=stratml.us&sitesearch=stratml.us>
> of the StratML collection turns up about 29 hits on the terms "AI
> framework".  Here's <https://www.modzy.com/platform-and-marketplace/> the
> top paid ad-placed hit (not yet in the StratML collection but soon to be).
>
thanks -  how do we query for ML explainability framework (a bit more
precise semantically in relation to what we are doing here)

KRID -  Carl is putting forward a category/concept/type  whereby KR is
identified
so KRID = some value to describe KR identity
Carl started by suggesting the top level distinction for this concept would
be
declarative/procedural
i  do not yet have an opinion about this, but would request Carl to start
sketching out
the taxonomy for KRID as he envisions it. so that we can have a discussion
about it
One considertation is: to what extent is declarative/procedural knowledge
relevant to support ML?  or is KRID intended for AI in general (not ML) .
Carl perhaps you should create this as a goal for yourself.Also could you
clarify the relation of KRID to KAIROS?

Thanks!

PDM
Received on Thursday, 14 May 2020 11:37:06 UTC

This archive was generated by hypermail 2.4.0 : Thursday, 14 May 2020 11:37:07 UTC