TensorFlow vs. FBP

[Paul Tarvydas]

> Indeed. Maybe a discussion of TensorFlow vs. FBP would be useful some
time as well (in another thread perhaps).
> It seems pretty data flow-ish, albeit unfortunately not incorporating
useful FBP stuff like ports :/

It appears to me that tf.placeholder() is much like a port.

>
> Also, it surprises me a bit they are not using Go more, as it is a
Google language and would give them a lot of data flow / fbp-features
out of the box.

Tensorflow 1.0 appears to have a Go directory. I doubt that go channels
can be used in a tf graph, though.

pt

[Paul Tarvydas]

Caveat: my suggestion that FBP'ers should think about TensorFlow is a "question" not a "conclusion".  This might turn out to be a blind alley.

Tensors were invented in physics, back when FORTRAN was still hot and OO was in its infancy.  Tensors are multi-dimensional arrays of Objects, where the objects can be ints, complex, 3d vectors, etc.

Fields are not uniform in space.  Imagine two waves in water colliding.  The interference patterns form hills and valleys on the surface of the water.  Think of describing such a complicated sets of field interactions in 3D space, using vectors.  At one point in the field, a 3D vector may describe the direction of the field (energy), say (3,4,5).  At the next point over, the energy interactions may result in a 3D vector (1,5,4).

This video (12 min) was most helpful in reminding me what tensors are.
https://www.youtube.com/watch?v=f5liqUk0ZTw

I found this lecture, albeit long, to give me a good sense for what TensorFlow can do
https://www.youtube.com/watch?v=vq2nnJ4g6N0

The slides seem to be here:
https://cloud.google.com/blog/big-data/2017/01/learn-tensorflow-and-deep-learning-without-a-phd

pt

[Paul Tarvydas]

GTA'ers (Greater Toronto Area) keep your eyes on this page. Next
seminar March 16 (registration is free). [Geoff Hinton has been there
the last two times I attended].

http://www.fields.utoronto.ca/activities/16-17/machine-learning

pt

[Samuel Lampa]

On Tuesday, March 7, 2017 at 6:08:04 PM UTC+1, Paul Tarvydas wrote:

This video (12 min) was most helpful in reminding me what tensors are.
https://www.youtube.com/watch?v=f5liqUk0ZTw

That was a great one! :)

I also found this O'Reilly Radar post interesting, which was published some time before TensorFlow was released, IIRC:
https://www.oreilly.com/ideas/lets-build-open-source-tensor-libraries-for-data-science

Will have to schedule some time for that longer video too :)

// Samuel