[x3d-public] My paraphrased version of AI history and its future with X3D XML and X3D JSON

[John Carlson]

You can skip 1 paragraph to move to future.

First there was conceptual dependency.   10-12 high level actions that described everything in the world.
Then were templated scripts.  Collections of actions in bundled in objects called scripts.
Then there were actors.   Separate threads of control which communicated with each other.
Then there were real networks, and the protocols became the templates we filled our data with.
Now there are software defined networks (SDNs), aka neural networks.   The protocol is very simple in this case, merely floating point integers or categorical variables.  Recently, this has been expanded to image (CNNs) and textual data (RNNs).

What will we see in the future?   Software defined neural networks which communicate objects (object2vec from amazon), data structures, templated scripts, actors, networks, SDNs.

Where am I in all of this?  I am working towards communicating XML and JSON (data structures) between neural networks, in particular creating test cases for communicating generative adversarial networks GANs). In the adversarial part, I would like to include rating systems of different communicating renderers and adversaries.  So someone could send a X3D XML or X3D JSON to a system, and it would render the data structure without explicitly coding an X3D browser.  The neural network would output an image, and validators would grade the rendering.  I think we only need a success/failure for the validating the rendering.   We would also be creating validators based on feedback from authoritative sources, and we would be creating more successful validators (for humans and systems who are writing X3D code still).

So I think at a high level, we want to create an RNN to analyze the text to a reverse-CNN to produce an image.  So we might need a CNN to take in an image and produce XML or JSON as well

Does anyone want to join me on the adventure? I am looking for a neural network which communicates data structures like XML and JSON. I am not sure we can communicate actors in an SDN, and I will have to think about how to communicate actors.

I am also open to doing something similar with the Universal Scene Description format.   Thanks!

We are presently creating a Python API for generating Python, so your Python Fu would be most welcome!  Then we should be able to manipulate X3D data structures with Python.  Woohoo!

Should I look through Kaggle competitions?  Hmm

John

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://web3d.org/pipermail/x3d-public_web3d.org/attachments/20190407/c989d2f3/attachment.html>