• Log in with Facebook Log in with Twitter Log In with Google      Sign In    
  • Create Account
  LongeCity
              Advocacy & Research for Unlimited Lifespans


Adverts help to support the work of this non-profit organisation. To go ad-free join as a Member.


Photo

Neural Network backpropagation?


  • Please log in to reply
5 replies to this topic

#1 olaf.larsson

  • Guest
  • 583 posts
  • 21
  • Location:Sweden

Posted 14 November 2006 - 07:05 AM


I have not yet fully understood neural network backpropagation could someone please explain it for me, in a farly easy way?

Edited by olarsson, 14 February 2007 - 06:58 PM.


#2 kgmax

  • Guest
  • 75 posts
  • 0

Posted 17 November 2006 - 12:04 AM

Thanks alot for bringing up this topic (sarcasm) ... well not totally sarcastic.

How much do you know about nueral nets and reinforcement learning?

I can point you to some links.

http://www.generatio...Neural Networks

http://www.cs.ualber...k/the-book.html

http://www.aaai.org/...html/reinf.html

http://ocw.mit.edu/N...uralNet2002.pdf

I apologize that I cannot explain it in simple terms. I could try but I am afraid it would make me look stupid.

sponsored ad

  • Advert

#3 olaf.larsson

  • Topic Starter
  • Guest
  • 583 posts
  • 21
  • Location:Sweden

Posted 18 November 2006 - 03:55 PM

Thank you for the links.

#4

  • Lurker
  • 0

Posted 16 January 2007 - 01:28 AM

Backpropagation is just a numerical optimization technique - nothing particularly deep or mysterious; it is a numerical optimization technique that attempts to find a local minima of the error surface of the neural network. There are much better methods for fitting neural networks.

Personally, I am not particularly fond of neural networks. They have their place but I think the research in machine learning/statistics has largely moved beyond them. They were hot back in the late 80's and throughout the 90's but I think there was a bit of irrational exuberance surrounding them.

#5 olaf.larsson

  • Topic Starter
  • Guest
  • 583 posts
  • 21
  • Location:Sweden

Posted 17 January 2007 - 09:06 AM

You seem very well informed about this subject ludongbin. So what is do you consider 'hot' in machine learning today?

sponsored ad

  • Advert

#6

  • Lurker
  • 0

Posted 18 January 2007 - 04:55 AM

> You seem very well informed about this subject ludongbin. So what is do you consider 'hot' in machine learning today?

Well, pretty much anything Bayesian is hot these days...but not unlike neural networks, some people get carried away. On the frequentist (i.e. non-Bayesian) side of things, SVMs (support vector machines) are pretty neat - has a lot of nice theory behind it and is based on solid mathematics (empirical process theory, etc.) SVMs are appealing in part because the optimization problem that needs to be solved to fit them is "nice" (unlike neural networks were there are lots of local minima, etc.) But I am not really working in this area right now so there are probably other more recent things happening as well.




1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users