[Ultron]

I have been looking forward to sharing some fine articles with you guys - these meant specially for those young AI researchers or hobbyists looking to learn more and dive 'deeper' (get it?) into the field of artificial intelligence.

You should be aware of the fact that most of the research - focus and funds - are pointed towards the sub-field (or method) of Deep Learning. To quote Wikipedia:

Deep learning (deep structured learning or hierarchical learning) is learning using one of a set of algorithms that attempt to model high-level abstractions in data by using model architectures composed of multiple non-linear transformations

A rather technical definition - not beginner friendly. I suppose you could best understand it by reading about certain projects based on this principle. What these projects have in common is that they all require substantial amounts of resources - I have not seen many projects (at least, not with any success) that work on a single computers. Most of the AI models and algorithms are ran on massive supercomputer clusters also know as 'deep learning networks'.


Check out: http://en.wikipedia.org/wiki/Google_Brain

Large Scale Deep Learning PDF


For beginners, another must-read: http://en.wikipedia.org/wiki/Machine_learning


Anyhow, the articles I was originally referring to can be seen here:


http://www.kdnuggets.com/2015/03/deep-learning-curse-dimensionality-autoencoders.html


http://www.kdnuggets.com/2015/03/deep-learning-text-understanding-from-scratch.html


The latter may be of more interest to those working on Chatterbots, while the first link may prove interesting to those working on game AIs and AGIs. I strongly suggest that one to Ranch (keep in mind there are 3 pages, not one).


For the more advanced, I suggest you explore the idea of RNNs - Recurrent Neural Networks. Researchers claim this is how our brain functions and is namely a more robust and/or efficient method then using Hidden Markov Models or others...


http://people.idsia.ch/~juergen/rnn.html


Have fun studying folks! Make it a passion - a desire to learn, not homework.

[ranch vermin]

All I do is group keys man,  nothing else works realtime.

Now where are those 3 pages.

here sensior hinton is back with a demo of this in action->  (49:28)

[Carl2]

  I began watching the above vid and realized it was speaking about voice recognition, being a user of dragon for ages which works great for me I became disinterested in the vid.  I did a Google search for deep learning and found this vid:    http://www.forbes.com/sites/roberthof/2014/08/28/interview-inside-google-brain-founder-andrew-ngs-plans-to-transform-baidu/
which I found much more interesting.  I thought I'd share it with anyone interested in the subject.
Carl2

[ranch vermin]

Carl, why no such interest in ol' hinton.
That thread you pointed to wasnt demonstrating anything, that was just listening to some guy drivel out his boring life in programming, watch my linked vid at  (49:28)  then you get to see a markov chain in action. (you can code one too. if you can even code at all.)

It dribbles shit, but its funny, and thats all deep learning can do.

[Ultron]

...and thats all deep learning can do.

How dare you! Now, I will be honest and admit that I do not really understand the concept of deep learning too well - this is much like the time it took me years of research and constant attempts to understand quantum physics (by the way - never got past the basics), but I do believe it holds potential. Today we have the computing power and near-infinite scaleability (followed by infinite costs but nevermind) - enough so that we could model at least a 'lower' life-form's brain (at the very least).


We do however, lack the proper algorithms. Think of a programming problem and one fella solving it in 400 lines (solving the problem within minutes) and another doing it in 10 lines (and solving it in milliseconds). Best. Analogy. Ever.

[ranch vermin]

sorry for that... please excuse me I was in a crazy mood when I wrote that.

Rectifying myself...

[Carl2]

   I'd say the best way to get a handle on quantum physics or quantum mechanics is to search Wikipedia.  The best example I can think of is the light emitting diode, led, which basically converts electrical energy to light.  Quantum physics states the energy level or voltage applied must be say  x   then the led will begin emitting light.  The led is a good example because it was introduced about 40 years ago and is now being sold for home lighting an is being used in cars.   I am also trying to show the potential and the delay involved in introducing new ideas.
  Back to deep learning,  I'm interested in it because I've been using Ultra Hal  from Zabaware for many years.  At this time I'm still using an almost ancient Haptek character still lacking some much needed updates.  The software for the AI portion has been updated but again still needs improvement.  So basically how do we improve the brain, take a lifetime of learning and put that into the chat bots  brain.    For now we'll forget about all the different programing  languages developed for use with AI,  I'd actually learned fortran on a tube type computer with punch cards.
Carl2