Thread: Some starting advice...

  1. #1
    Registered User
    Join Date
    Nov 2004
    Location
    Pennsylvania
    Posts
    434

    Some starting advice...

    Hello everyone, well recently i've become entranced in this book entitled: AI, The Tumoltuous History of Artificial Intelligence by Daniel Crevier. If you haven't read it you should, its a great book, very unique in its approach.

    Well anyways, me being the person i am as far as jumping from one thing to another...

    Are there any simple approachs to some AI topics, such as neural networks, that i could get started on? Or perhaps another book which covers a more technical approach?


    Thanks!
    "Anyone can aspire to greatness if they try hard enough."
    - Me

  2. #2
    Officially An Architect brewbuck's Avatar
    Join Date
    Mar 2007
    Location
    Portland, OR
    Posts
    7,396
    Quote Originally Posted by Junior89 View Post
    Are there any simple approachs to some AI topics, such as neural networks, that i could get started on? Or perhaps another book which covers a more technical approach? Thanks!
    Honestly, I think neural networks are a bad place to start. I'd start with some simple search-based stuff and a background on predicate calculus. Try "Artificial Intelligence: A New Synthesis" by Nils Nilsson.

    Neural networks are really a very powerful nonlinear interpolation technique for high-dimensional functions. They were inspired by biological neural networks, so they were originally considered "AI" but the traditional methods like backprop are pretty much nuts and bolts these days. The cool stuff is happening elsewhere. Nilsson's book does cover them, though, so you can read about them there.

  3. #3
    Registered User
    Join Date
    Nov 2004
    Location
    Pennsylvania
    Posts
    434
    Thanks!
    "Anyone can aspire to greatness if they try hard enough."
    - Me

  4. #4
    Crazy Fool Perspective's Avatar
    Join Date
    Jan 2003
    Location
    Canada
    Posts
    2,640
    > Neural networks are really a very powerful nonlinear interpolation technique for high-dimensional functions.

    Actually the simplest case of a Perceptron is a linear example, but I agree with your post anyway. Start with some simple search algos, decision trees, etc...

  5. #5
    Officially An Architect brewbuck's Avatar
    Join Date
    Mar 2007
    Location
    Portland, OR
    Posts
    7,396
    Quote Originally Posted by Perspective View Post
    Actually the simplest case of a Perceptron is a linear example, but I agree with your post anyway.
    Only a single-layer perceptron. A linear, multi-layer perceptron is equivalent to a single-layer perceptron. Therefore there is no point in creating a multi-layer, linear perceptron. So all multi-layer perceptrons are by definition nonlinear.

  6. #6
    Crazy Fool Perspective's Avatar
    Join Date
    Jan 2003
    Location
    Canada
    Posts
    2,640
    Actually, it's the perceptron itself that is linear. When you move to something non-linear (eg. sigmoid) it's no longer a perceptron.

    You're right that there's no point in having multiple layers as they are always equivalent to some single layer perceptron though, but that's a different point.

  7. #7
    Registered User
    Join Date
    Nov 2004
    Location
    Pennsylvania
    Posts
    434
    Perceptrons... haha, i read about them too i think. =P

    But i'll start small Like search algorithms, decision trees, etc.

    And work my way up =)
    "Anyone can aspire to greatness if they try hard enough."
    - Me

  8. #8
    Officially An Architect brewbuck's Avatar
    Join Date
    Mar 2007
    Location
    Portland, OR
    Posts
    7,396
    Quote Originally Posted by Perspective View Post
    Actually, it's the perceptron itself that is linear. When you move to something non-linear (eg. sigmoid) it's no longer a perceptron.

    You're right that there's no point in having multiple layers as they are always equivalent to some single layer perceptron though, but that's a different point.
    Despite that, the term "multi-layer perceptron" continues to be used for multi-layer NONLINEAR networks. See this:

    http://en.wikipedia.org/wiki/Artific...yer_perceptron

    This class of networks consists of multiple layers of computational units, usually interconnected in a feed-forward way. Each neuron in one layer has directed connections to the neurons of the subsequent layer. In many applications the units of these networks apply a sigmoid function as an activation function.

  9. #9
    Crazy Fool Perspective's Avatar
    Join Date
    Jan 2003
    Location
    Canada
    Posts
    2,640
    Interesting, I've never heard the term perceptron as meaning something non-linear. Though wikipedia does contradict itself with the definition of a perceptron:

    http://en.wikipedia.org/wiki/Perceptron

    The perceptron is a type of artificial neural network invented in 1957 at the Cornell Aeronautical Laboratory by Frank Rosenblatt. It can be seen as the simplest kind of feedforward neural network: a linear classifier.


    The whole motivation behind multi-layer non-linear networks is that perceptrons can only solve (classify) linearly seperable problems.

    But anyway, it's just a terminology issue.

  10. #10
    Officially An Architect brewbuck's Avatar
    Join Date
    Mar 2007
    Location
    Portland, OR
    Posts
    7,396
    Quote Originally Posted by Perspective View Post
    Interesting, I've never heard the term perceptron as meaning something non-linear. Though wikipedia does contradict itself with the definition of a perceptron:
    I don't think it's a contradiction, I think it is bad terminology which Wikipedia didn't invent. A "multi-layer perceptron" really isn't a derived class of "perceptron," it's a whole other thing. Somebody probably misspoke or used bad terminology at some point and the nomenclature stuck.

Popular pages Recent additions subscribe to a feed

Similar Threads

  1. Interpreter.c
    By moussa in forum C Programming
    Replies: 4
    Last Post: 05-28-2008, 05:59 PM
  2. help a n00b starting out. any advice?
    By MegaManZZ in forum C++ Programming
    Replies: 5
    Last Post: 11-30-2007, 02:10 AM
  3. girl friend advice (prob. the wrong place)
    By B0bDole in forum A Brief History of Cprogramming.com
    Replies: 15
    Last Post: 10-22-2004, 06:38 PM
  4. starting C++ advice
    By bobbob in forum C++ Programming
    Replies: 2
    Last Post: 10-21-2003, 12:10 PM
  5. advice on starting band?
    By Waldo2k2 in forum A Brief History of Cprogramming.com
    Replies: 9
    Last Post: 11-10-2002, 01:05 PM