Do Bayesian statistics rule the brain?:

Get Feed

Description

This week's New Scientist has a fascinating article on a possible 'grand theory' of the brain that suggests that virtually all brain functions can be modelled with Bayesian statistics - an approach discovered by an 18th century vicar.

Original URL

http://www.mindhacks.com/blog/2008/05/do_bayesian_statisti.html

http://www.mindhacks.com/blog/2008/05/do_bayesian_statisti.html

Comments

• Public Comments

  • johnrod

    18 months ago

    
    
    

    James: Is this making the case that the semantic web will be Bayesian in nature (beyond conditional probabilities, data-mining and inference engines)? Thanks.

    James: Is this making the case that the semantic web will be Bayesian in nature (beyond conditional probabilities, data-mining and inference engines)? Thanks.

    • Some HTML is allowed.
    • Reply
    • Cancel
    • Submit
    Commented in: Solar Soyuz Zaibatsu

    • James Choate

      18 months ago

      
      
      

      Well let's start out by making it clearer about what it is we are talking about...
      
      In normal probability we take some population of events and say that for a particular outcome we have some probability x that it will occur. In Bayesian statistics we ask does a particular event/sample fit a model to some degree of accuracy (eg standard deviation). The actual data is no different, it's the kind of question we ask.
      
      Now to get to your question, it seems that Bayesian statistics are more ameniable to neural networks because of the architecture and the way neural network elements 'ask' questions on their input. So, I would have to say that in some cases Bayesian filters are a better way to look at a pile of data and make a guess as to what is really important there. All three of the things you mentioned can be done with either traditional population based (ie Laplacian) or Bayesian statistics. It's more how you want to look at it. More to the point I think it's stating that because of the apparent way brains work we'll find that the best tools for digging through large piles of data will be Bayesian in nature at well. The web as a whole will have components of both perspectives, depending on the kinds of questions you're asking.
      
      I think what it is saying is you need to be able to deal with both approaches in the real world :)

      Well let's start out by making it clearer about what it is we are talking about...<br /><br />In normal probability we take some population of events and say that for a particular outcome we have some probability x that it will occur. In Bayesian statistics we ask does a particular event/sample fit a model to some degree of accuracy (eg standard deviation). The actual data is no different, it's the kind of question we ask.<br /><br />Now to get to your question, it seems that Bayesian statistics are more ameniable to neural networks because of the architecture and the way neural network elements 'ask' questions on their input. So, I would have to say that in some cases Bayesian filters are a better way to look at a pile of data and make a guess as to what is really important there. All three of the things you mentioned can be done with either traditional population based (ie Laplacian) or Bayesian statistics. It's more how you want to look at it. More to the point I think it's stating that because of the apparent way brains work we'll find that the best tools for digging through large piles of data will be Bayesian in nature at well. The web as a whole will have components of both perspectives, depending on the kinds of questions you're asking.<br /><br />I think what it is saying is you need to be able to deal with both approaches in the real world :)

      • Some HTML is allowed.
      • Reply
      • Cancel
      • Submit
      Commented in: Solar Soyuz Zaibatsu

  • Nova Spivack

    18 months ago

    
    
    

    Very interesting. I wonder if Bayesian statistics also rule quantum mechanics and General Relativity. Could all our physical laws be reduced to Bayesian systems? I have a hunch that gravity could be expressed as a "Bayesian force" or a "Bayesian field."

    Very interesting. I wonder if Bayesian statistics also rule quantum mechanics and General Relativity. Could all our physical laws be reduced to Bayesian systems? I have a hunch that gravity could be expressed as a "Bayesian force" or a "Bayesian field."

    • Some HTML is allowed.
    • Reply
    • Cancel
    • Submit
    Commented in: Solar Soyuz Zaibatsu

  • James Choate

    18 months ago

    
    
    

    Well strictly speaking Bayesian and Laplacian statistics are equivalent, it's really more a difference in the way you pose the question. I think of Laplacian as a definitive/closed perspective and Bayesian a more objective/open perspective. Where the beauty of Bayesian comes in is because the difference in the way the question is posed changes the way you build your model. Bayesian for larger data sets seem to be easier to program and handle larger data sets (ie O() isn't as bad).

    Well strictly speaking Bayesian and Laplacian statistics are equivalent, it's really more a difference in the way you pose the question. I think of Laplacian as a definitive/closed perspective and Bayesian a more objective/open perspective. Where the beauty of Bayesian comes in is because the difference in the way the question is posed changes the way you build your model. Bayesian for larger data sets seem to be easier to program and handle larger data sets (ie O() isn't as bad).

    • Some HTML is allowed.
    • Reply
    • Cancel
    • Submit
    Commented in: Solar Soyuz Zaibatsu

  • 18 months ago

    
    
    

    That might be how the neurological substrates work. But normal human thinking isn't Bayesian. Check out the Monty Hall problem here: http://en.wikipedia.org/wiki/Monty_Hall_problem.

    That might be how the neurological substrates work. But normal human thinking isn't Bayesian. Check out the Monty Hall problem here: http://en.wikipedia.org/wiki/Monty_Hall_problem.

    • Some HTML is allowed.
    • Reply
    • Cancel
    • Submit
    Commented in: Solar Soyuz Zaibatsu

    • James Choate

      18 months ago

      
      
      

      People in general think in allegory or by 'type' which is certainly Bayesian in -many- ways.
      
      But to your point, nobody is making that claim.

      People in general think in allegory or by 'type' which is certainly Bayesian in -many- ways.<br /><br />But to your point, nobody is making that claim.

      • Some HTML is allowed.
      • Reply
      • Cancel
      • Submit
      Commented in: Solar Soyuz Zaibatsu

  • Kurt Laitner

    18 months ago

    
    
    

    check the notes on solar soyuz zaibatsu twine

    check the notes on solar soyuz zaibatsu twine

    • Some HTML is allowed.
    • Reply
    • Cancel
    • Submit
    Commented in: Structure

  • Spaceweaver

    14 months ago

    
    
    

    You would certainly be interested in the work of Jeff Hawkins and the book On Intelligence. Will share the items to this twine.

    You would certainly be interested in the work of Jeff Hawkins and the book On Intelligence. Will share the items to this twine.

    • Some HTML is allowed.
    • Reply
    • Cancel
    • Submit
    Commented in: Advanced Computing :: Neural Networks, Fuzzy Systems, Genetic Algorithms & Evolutionary Computation
  Add a Comment

Here's a list of the HTML tags we allow: