[Chicago] Python for AI?

[Jeremy McMillan]

I'd like to add that the reason functional programming is popular for AI is
because Artificial Neural Networks (ANNs) are made of (RIP Marvin Minsky)
independent agents which perform similar, computationally cheap operations
on many inputs, reduced to a single output.

Functions are a tidy way to encapsulate this pattern. In this style, if a
function is defined to operate on multiple iterations of factors, and
produce a vector of output values, then it can simulate any neuron of a
certain type (activation operation/output). Usually the operation is to
multiply the input values from connected neurons each (hello map()!) by a
weight state for each connection, and then reduce those products to a
statistical mean, or htan() of the mean. Can you see the map/reduce
pattern? Matrix multiplication also works for certain types of neurons. Can
you think of good object oriented approaches?

I highly recommend the seminal Coursera Machine Learning course, though it
is taught in Matlab/Octave.

Also check out Geoff Hinton's 2007 Google Talk 'The Next Generation of
Neural Networks' for an introduction to Deep Learning (example code is also
Matlab).
https://youtu.be/AyzOUbkUf3M

After that, head over to http://deeplearning.net
Python is there!

On Tue, Jan 26, 2016, 23:08 Joshua Herman <zitterbewegung at gmail.com> wrote:

> Dear Lewit,
> Functional programming isn't important in AI. There are plenty of AI
> programs that are programmed in imperative style. It just happened
> that Lisp was a functional language and people used it to implement AI
> concepts. Right now it seems like most scientists are using either R,
> Python or Matlab , Also, in deep learning a bunch of the libraries are
> implemented in Python .
> Sincerely,
> Joshua herman
>
>
> On Tue, Jan 26, 2016 at 9:55 PM, Lewit, Douglas <d-lewit at neiu.edu> wrote:
> > Thanks Thomas.  I'll look into these links.  I know that AI specialists
> are
> > attracted to functional programming.  Lisp is the oldest of the
> functional
> > languages, but I believe Python is pretty flexible because it supports
> > different paradigms, including the functional paradigm.  Of course....
> the
> > next question is, "Why is functional programming so important in AI?"
> >
> > On Tue, Jan 26, 2016 at 5:09 PM, Thomas Johnson <
> thomas.j.johnson at gmail.com>
> > wrote:
> >>
> >> I think perhaps your prof is not covering modern history in machine
> >> learning. I don't think LISP is very common in AI anymore. The ML world
> is
> >> dominated mostly by Python, R, and Matlab as high-level languages and
> C++
> >> for low-level work (with maybe some FORTRAN for really hard-core math
> >> routines).
> >>
> >> In fact, I would say Python is probably one of the leading AI languages,
> >> probably even the top language when combined with C++ (although there's
> a
> >> lot of cutting-edge statistical stuff that's still only available in R
> or
> >> Matlab). For example, Theano (http://deeplearning.net/software/theano/)
> is a
> >> widely-used library for Deep Learning, and Google's new TensorFlow
> >> (https://www.tensorflow.org/) has bindings for C++ and Python. Plus of
> >> course there's Scikit-Learn (http://scikit-learn.org/). There's even a
> whole
> >> python distribution (https://www.continuum.io/why-anaconda) whose sole
> >> purpose is to get you set up quickly for data science work in Python.
> >>
> >> That's just a handful of the tools available in Python
> >>
> >> On Tue, Jan 26, 2016 at 4:59 PM Lewit, Douglas <d-lewit at neiu.edu>
> wrote:
> >>>
> >>> Hi everyone,
> >>>
> >>> I'm taking an AI class at Northeastern.  Interesting class, but sadly
> no
> >>> programming in the class.  More a survey course covering the history
> and
> >>> some of the algorithms of AI.  However, the prof wants us to write a
> paper
> >>> on AI that's due some time in April.  We're free to choose whatever
> topic we
> >>> want, as long as it's related to AI.  Something that might be fun to
> write
> >>> about is Python for AI.... maybe?  I know that LISP is usually the
> language
> >>> of choice for AI research, and I've been playing around with Lisp
> lately
> >>> using the Clozure CL IDE.  Definitely a nice language, although it
> will take
> >>> me a couple more days to get comfortable with Lisp's prefix notation.
> But
> >>> what about Python?  Does anyone on this list have any experience with
> Python
> >>> for AI?  Can you recommend any books or websites or articles?  If so,
> please
> >>> let me know.
> >>>
> >>> Thanks for your help!
> >>>
> >>> Best,
> >>>
> >>> Douglas.
> >>>
> >>> _______________________________________________
> >>> Chicago mailing list
> >>> Chicago at python.org
> >>> https://mail.python.org/mailman/listinfo/chicago
> >>
> >>
> >> _______________________________________________
> >> Chicago mailing list
> >> Chicago at python.org
> >> https://mail.python.org/mailman/listinfo/chicago
> >>
> >
> >
> > _______________________________________________
> > Chicago mailing list
> > Chicago at python.org
> > https://mail.python.org/mailman/listinfo/chicago
> >
> _______________________________________________
> Chicago mailing list
> Chicago at python.org
> https://mail.python.org/mailman/listinfo/chicago
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/chicago/attachments/20160127/8e70ac96/attachment.html>