[x3d-public] Declarative Bayesian approach to X3D (currently using randomnumbergenerator in JavaScript) in X3DOM and X_ITE.

[John Carlson]

It seem like Berners-Lee biggest issue may have been the update of probabilities.  Hmm.

If I had a brain fart and deep learning came out (ONNX http://onnx.ai/ —Inputs=Outputs=nodes,attributes,relationships,weights,probabilities=semantic web), that would be great.

John

Sent from Mail for Windows 10

From: John Carlson
Sent: Friday, February 8, 2019 11:01 AM
To: X3D Graphics public mailing list; Don Brutzman; Leslie Sikos
Subject: RE: Declarative Bayesian approach to X3D (currently using randomnumbergenerator in JavaScript) in X3DOM and X_ITE.

I am not sure or not, but I think Tim Berners-Lee may have said the semantic web does not include probabilities.   Do I need to create my own “relationships between nodes” and “relationships between nodes and attributes” in order to add probabilities? Go ahead and use XMLBIF?

Sent from Mail for Windows 10

From: John Carlson
Sent: Friday, February 8, 2019 10:39 AM
To: X3D Graphics public mailing list; Don Brutzman; Leslie Sikos
Subject: Declarative Bayesian approach to X3D (currently using random numbergenerator in JavaScript) in X3DOM and X_ITE.

My understanding of Bayes systems is they make use of a network of conditional probabilities.  The ideas below refer to a Bayesian approach to “novel” or test data XML generation from the root of the document hierarchically out to the leaves with probabilities for nodes, relationships and attributes appearing in an ontology, because I have no other current place to put it. Other suggestions are welcome.   Thus I am making advance to the Ontology working group to provide for probabilities in the OWL/RDF.  There may be a better place for the probabilities (discuss), see Emitter nodes.  Also see: https://www.quora.com/Theoretical-Physics-Is-probability-epistemological-or-ontological

So I am looking for elements (in any XML language, but preferably X3D) which apply Bayesian approaches.
For example, I might use Bayesian methods on pixels to create movies, textures, sounds and volume rendering.
I might use Bayesian methods on text (natural language generation) to create speech, rendered text, comments, and identifiers
I might use Bayesian methods on sound encoded with vectors, vectors, surfaces, volumes (motion capture) and to animate those things (and higher dimensions).  See my flowers, force.
I might use Bayesian methods to apply a small amount of variation to an HAnim skeleton animation.
I might use Bayesian methods as described in the paragraph above.
Bayesian parameters could be applied directly to ROUTEs for example, and I think is appropriate there, such that events fire based on a random variation, and is the proportion of the total number of times the ROUTE should fire in normal X3D.

I see the Emitters (Particle Systems) have variation. How can I apply that to non-Cartesian coordinate systems, and how do I feed that into a dynamic sound? I will do some research. Are Emitters present in current browsers?  Thanks!

Should I form a small working group on to create a wiki page to document the Declarative Bayesian methods and X3D?  Or have we been there, done that?

I discovered this:   http://www.cs.cmu.edu/afs/cs/user/fgcozman/www/Research/InterchangeFormat/  (version 0.3) How do I go about integrating X3D with XMLBIF?
Also, if anyone has any other primitive visual, sound and other senses (tactile/haptic) data format other than pixels, vectors and text, let me know.

How might the Bayesian working group interact with the Ontology working group. Can the Ontology group provides us with the needed parameters, priors and probabilities? One of the objectives of the Semantic working group is: automatic generating of X3D documents based on ontologies and vice versa, http://www.web3d.org/working-groups/x3d-semantic-web/charter  Are these OWL/RDF documents, X3D XML documents?   Are we attempting to generate novel documents, or a mapping (then use “mapping”).  If we are creating novel documents, may I suggest that Bayesian techniques might be used?   What other techniques for creating document.

How can we collect data for a Bayesian generated document?  Use various mapping databases?   Google Earth? OGC?

So you can think of Bayesian methods as a Bayesian network tied in space to data.  Bayes methods can be thought of as N-dimensional
.
The intention is to provide *pleasant* not necessarily *realistic* visuals, something like white noise could be pleasant, for example for generating audio (a stream of water). 

The intent is to animate the scene in real-time, rather than providing fixed animation, images, text, drawings and sounds.

Other ideas for Bayesian, Brownian and Chaotic methods are welcome.


Thanks,

John


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://web3d.org/pipermail/x3d-public_web3d.org/attachments/20190208/6e349a36/attachment-0001.html>