nov 12

homework

• 13.12
• 13.15
• 13.16

Jim's Bayes spam classification - results

• Jim spam folder
• bayes.pl especially "sub bayes_from_file"
• output 1
• output 2

neural nets

• see pgs 736 and following in text

• brain ( http://www.willamette.edu/~gorr/classes/cs449/brain.html | local )

• artificial neuron ( http://www.willamette.edu/~gorr/classes/cs449/ann-overview.html | local )

• networks are
  • feed-forward (i.e. signal propogates from input to output)
    • well studied
    • well understood, though math gets involved
    • trainable perception systems
  • recurrent (i.e. signal propogates in loops; outputs feed back to inputs)
    • can have memory (i.e. current state depends on past history)
    • closer to model of brain
    • much harder to understand
    • but see
      • http://www.shef.ac.uk/psychology/gurney/notes/l5/l5.html
      • http://en.wikipedia.org/wiki/Hopfield_net

• Network implementation
  • Weights(i,j) are numbers that describe connection strength from node i to node j.
  • nonlinear threshold function g()
  • output of node i is

• Training
  • typically weights are determined by backpropogation algorithm
  • backprop explanation ( http://galaxy.agh.edu.pl/~vlsi/AI/backp_t_en/backprop.html | local)

• digit recognition example
  • http://lslwww.epfl.ch/~anperez/NN_tutorial/NNtutorial.html | local

final projects

• discuss