Got a YouTube account?

New: enable viewer-created translations and captions on your YouTube channel!

English subtitles

← 03_Induction and Deduction

Get Embed Code
2 Languages

Showing Revision 1 created 01/18/2014 by Cogi-Admin.

  1. Well, what you're doing in order to make that work, and
  2. what you end up doing in supervised learning and function approximation
  3. in general, is you make some fundamental assumptions about the world.
  4. Right? You decide that you have a well-behaved function that is
  5. consistent with the data that you're getting. And with that, you're
  6. able to generalize. And, in fact, that is the fundamental problem
  7. in Machine Learning. It is generalization. Now what's behind all of
  8. this, I'm going to claim. Michael, you jump in whenever you disagree.
  9. >> I disagree! Oh,
  10. sorry. Too soon. Go ahead.
  11. >> Is bias.
  12. >> Bias.
  13. >> And in particular, inductive bias.
  14. >> Inductive bias.
  15. >> Right. So all of Machine Learning, or certainly
  16. Supervised Learning, is about induction. As opposed to deduction
  17. >> I see, induction of course being a problem
  18. of going from examples to a more general rule.
  19. >> Right specifics to generalities. By contrast deduction is?
  20. >> Would be the opposite. It would be
  21. going from a general rule to specific instances,
  22. basically like reasoning. Right, in fact a lot of the
  23. AI in the beginning was about productive reasoning about logic,
  24. programming, those sort of things where you have certain rules
  25. and you do only those things that follow mainly in those
  26. rules. For example you have A implies B, that's a
  27. rule in universe and then I tell you A so
  28. if you know A implies B in universe and I
  29. tell you A then you also know. That A implies b.
  30. >> And
  31. therefore, you can.
  32. >> And a.
  33. >> Infer that.
  34. >> B.
  35. >> B, you have a implies b, you have a. That implies b.
  36. >> Okay.
  37. >> It's what we just said. That's deduction.
  38. >> That's deduction but what we just did
  39. was not deduction. Before then when I asked you
  40. 1, 1, 2, 4, 3, 9, 4, 16 and so on and so forth, we did induction.
  41. >> That was induction.
  42. >> Induction is more about did the Sun rise yesterday?
  43. >> Yes.
  44. >> Did the Sun rise the day before that?
  45. >> Yes.
  46. >> Did the Sun rise the day before that?
  47. >> [UNKNOWN].
  48. Did the sun rise the day before that?
  49. >> Yes.
  50. >> Yes. So the sun has risen every day. Is the sun going to rise tomorrow?
  51. >> I sure hope so.
  52. >> We all hope so. And we all act
  53. like it does because if it doesn't, then there are
  54. a whole bunch of other things we ought to be
  55. doing besides sitting in this studio and having this interview.
  56. >> I think we should warn the plants.
  57. >> [LAUGH] I don't think the plants are going to care.
  58. >> They are! They really need sun.
  59. I think we all need some, Michael. So the idea there
  60. is that induction is crucial, and that inductive bias is crucial.
  61. And we'll talk about all this in the course. But that's
  62. kind of fundamental notions behind supervised
  63. learning and machine learning in general.
  64. >> I agree with that. Yeah.
  65. >> All right, so we're on the same page. So that's supervised learning.
  66. Supervised learning, you can talk about in these high muckety-muck ways. But at
  67. the end of the day It's function approximation. It's figuring out how to
  68. take a bunch of training examples, coming in with some function that generalizes
  69. beyond the data you see.
  70. >> So why wouldn't you call it function induction then?
  71. >> Because someone said supervised learning
  72. first. Well there is a difference.
  73. >> No, no, no, no you said supervised learning is
  74. function approximation and I want to
  75. say supervised learning is. Function induction?
  76. >> As opposed to function approximation?
  77. >> Yeah.
  78. >> Okay. It's
  79. >> Approximate function induction.
  80. >> Or induction of approximate, of.
  81. >> Approximate functions.
  82. >> Something like that, yeah.
  83. >> You don't want to induce
  84. an approximate function, you want to induce the actual function.
  85. >> Yeah, but sometimes you can't.
  86. >> Yeah.
  87. >> Because sometimes you think it's quadratic, but it's
  88. not. I have that as a plaque on my wall.
  89. >> You do?
  90. >> No.
  91. >> Yeah, I didn't think so. Okay, so that's supervised learning.