[semantics-public] RDF embeddings for X3D

[John Carlson]

Vince, embedding is a technical term from machine learning for taking large
amounts of text (word2vec) or objects (Amazon? Sagemaker—object2vec) and
converting word or object adjacency (in spatial or temporal terms) into
vectors.  This allows words (or things that otherwise aren’t reduced to
scalar or matrix values) to be used as vectors instead of using symbolic
manipulation.   One can even do vector arithmetic on words, and it works!
Obviously words farther away from a central word should probably have a
lower weight while embedding.

There are many …2vec projects.

For example one might use the 6 words around a word throughout a document
to predict what word comes next after completing embedding.

One needs a vector space to compute gradient descent in neural network
backpropagation, since gradient descent cannot be computed on written
words.  I believe that the embedding provides such a space, too.

In other words, embedding means converting data into vectors so things like
TensorFlow will work!

The old story used to be that a book can be converted into a vector of a
very large space.   Embedding is taking that huge space and reduces the
dimensions, and increases the number of vectors.

Hope this helps, and will eventually be another tool for 3D practitioners.

John

On Wed, May 25, 2022 at 8:52 PM vmarchetti at kshell.com <vmarchetti at kshell.com>
wrote:

> I do not know what capabilities Word2vec or SageMaker offer.
>
> I have been working on embedding XMP metadata packets in X3D Metadata
> nodes. XMP uses a subset of RDF. The conversions from XMP XML packets to
> and from X3D Metadata (XML Encoding) are implemented in XSLT stylesheet in
> the github project https://github.com/vincentmarchetti/x3d-xslt-tools .
>
> This work is in support of the ISO JWG-16 STEP Geometry Services project,
> which has adopted XMP as a schema for metadata to describe CAD web
> services. It may also be useful in embedding XMP in general in X3D .
>
> Vince Marchetti
>
> > On May 25, 2022, at 6:18 PM, John Carlson <yottzumm at gmail.com> wrote:
> >
> > Does anyone have RDF embeddings for X3D, like word embeddings for
> Word2vec or object embeddings for SageMaker?
> >
> > Thanks for info!
> >
> > John
> > --
> > semantics-public mailing list
> > semantics-public at web3d.org
> > http://web3d.org/mailman/listinfo/semantics-public_web3d.org
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://web3d.org/pipermail/semantics-public_web3d.org/attachments/20220525/5285f805/attachment.html>