Non-linear hypothesis

  |   Source
Non-linear hypothesis
  • Outdated but most powerful learning algorithm in most ML problems
  • the number of features is increasing as big theta notation
  • even only quadratic feature get picked,  only get circle,  not the more complex one
  • if not the the quadratic get picked,  the order will increase a lot more,  close to 170.000 features..
  • for many,  examples,  n tend to increase very large

  • examples of car image
  • thr computer vision is hard,  that's way
  • if we want to tell computer to recognize the image
  • the computer analyze the door knob
  • need to differentiate the cars and non cars
  • high computational expensive
  • n = 2500, for grayscale,  7500 for RGB,  3 million if quadratic included
  • to put it simply,  the logistic regression is expensive for lots of features
  • here's the neural networks actually works,  solving a non-linear problems  with a lot of complex features