YouTube

Got a YouTube account?

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

English subtitles

← 06_Comparison of these parts of ML

Get Embed Code
2 Languages

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

  1. >> Okay, so, we have got three little bits of a machine line
  2. here and there are lot of tools and techniques that are inside that.
  3. >> Mm-hm.
  4. >> And I think that's great and we are going to be
  5. trying to teach you a lot of those tools and techniques
  6. and sort of ways to connect them together. So, by the
  7. way, as Michael is pointing out, there are kind of ways
  8. that these might help each other. Unsupervised learning might help supervised
  9. learning, it's actually much deeper than that. It turns out you,
  10. even though unsupervised learning is clearly not the same as supervised
  11. learning, at the level we've described it, in some ways they're exactly
  12. the same thing. Supervised learning you have some bias, oh
  13. it's a quadratic function, induction makes sense, all these kind
  14. of assumptions you make. And in unsupervised learning, I told
  15. you that we don't know whether this cluster's better than
  16. this cluster, dividing by sex is better than dividing by
  17. height, or or hair color or whatever. But ultimately, you
  18. make some decision about how to cluster, and that means
  19. implicitly there's some assumed set. There's some assumed set of labels
  20. that you can possess. Oh, I think things that look alike should
  21. somehow be clustered. Things that are near one another should be clustered
  22. together. So, in some ways, it's still kind of like supervised learning. You
  23. can certainly turn any supervised learning
  24. problem into an unsupervised learning problem.
  25. >> Mm, mm.
  26. >> Right. So in fact, all of these
  27. problems are really the same kind of problem.
  28. >> Yeah, well there's two things that I'd
  29. want to add to that. One is that, in some
  30. sense, in many cases, you can formulate all
  31. of these different problems as some form of optimization.
  32. In supervised learning, you want something that, that labels data well, and
  33. so your, the thing you're trying to optimize is, find me a function
  34. that, that does that. We're going to score it. In reinforcement learning, we're
  35. trying to find a behavior that scores well. And in unsupervised learning, we
  36. usually have to make up some kind of a criterion, and then
  37. we find a way of clustering the data, organizing the data so that
  38. it scores well. So that was the first point I wanted to
  39. make. The other one is. If you divide things by sex and you're
  40. a virgin, then there's numerical instability issues.
  41. >> Did you learn about that on the street?
  42. >> I learned in the math book.
  43. >> That's ugh. I, I'm going to move on with her. So here's the thing.
  44. >> Alright.
  45. >> Everything Michael just said, except the last part, is
  46. true. But there's actually a sort of deeper thing going
  47. on, to me. If you think about the commonalities of
  48. everything we just said, it boils down to one thing, data.
  49. Data, data, data, data, data. Data is king in
  50. machine work. Now Michael would call himself a computer scientist.
  51. >> Oh, yeah.
  52. >> And I would call myself a computationalist.
  53. >> What?
  54. >> I'm in a college of computing at a Department
  55. of Computer Science. I believe in computing and computation as being
  56. the ultimate thing. So I'd call myself a computationalist and Michael
  57. would probably agree with that just to keep this discussion moving.
  58. >> Let's say.
  59. >> Right, so we're computationalists, we
  60. believe in computing. That's a good thing.
  61. >> Sure.
  62. >> Many of our colleagues, who do computation,
  63. think in terms of algorithms. They think in terms
  64. of, what are the series of steps I need to do in order to solve some problem?
  65. >> I think about algorithms.
  66. >> Or they might think in terms of theorems. If I try to
  67. describe this problem in a particular way,
  68. is it solvable implicitly by some algorithm?
  69. >> Yeah.
  70. >> And truthfully, machine learning is a lot
  71. of that. But the difference between the person who's
  72. trying to solve a problem as an AI
  73. person or as a computing person and someone who's
  74. trying to solve a problem as a machine learning person is
  75. that the algorithm stops being central, the data starts being central.
  76. And so what I hope you get out of this class
  77. or at least part of the stuff that you do, is understanding
  78. that you have to believe the data, you have to do
  79. something with the data, you have to be consistent with the data.
  80. The algorithms that fall out of all that are algorithms but
  81. they're algorithms that take the data as primary. Or at least important.
  82. >> I'm going to go with coequal.
  83. >> So algorithms and
  84. data are coequal.
  85. >> Coequal.
  86. >> Well, if you believe in lists they the same thing.
  87. >> Exactly.
  88. >> They knew back in the seventies.
  89. >> So it turns out we do agree on those things.
  90. >> Whew that was close.
  91. >> Excellent. So, the rest of the semester will go exactly
  92. like this except you won't see us. You'll see our hands though
  93. >> This side, this side. There you go.
  94. >> You'll see our hands though. Thank you Michael
  95. >> S'alright.
  96. >> [LAUGH] Well.
  97. >> What? [LAUGH]
  98. >> That was good. That took me back to when I was four.
  99. >> Señor, Señor Wences.
  100. >> Hm?
  101. >> It's called Señor Wences.
  102. >> Yes I know. I remember that.
  103. >> Yeah, okay. Mm-hm.
  104. >> I'm not that much younger than you. Ten, 12 years old.
  105. >> Come on.
  106. >> You count gray hairs. Anyway, the point is
  107. the rest of the semester will go like this.
  108. We will talk about supervised learning and a whole
  109. series of algorithms, step back a little bit and talk
  110. about the theory behind them. And try to connect theory of machine
  111. learning with theory of computing notions. Or at least that kind of
  112. basic idea. What does it mean to be a hard problem versus
  113. an easier problem. We'll move into
  114. randomized optimization and unsupervised learning. Where we
  115. will talk about all the issues that we brought up here and
  116. try to connect them back to some of the things that we
  117. did in the section on supervised learning. And then finally, we will
  118. spend our time on reinforcement learning
  119. and the generalization of these traditional reinforcement
  120. learning which involves multiple agents. So we'll talk about a little bit.
  121. >> Hm.
  122. >> Of game theory which Michael loves to talk about, I love
  123. to talk about. And the applications of all the stuff that we've
  124. been learning To solving problems. How to actually act in the world.
  125. How to build that robot to do something, or build that agent
  126. to play a game, or to teach you how to do whatever
  127. you need to be taught how to do. But at the end
  128. of the day we're going to teach you to think about data, how
  129. to think about algorithms, and how to build artifacts that you know
  130. will learn.
  131. >> Let's do this thing.
  132. >> Excellent. Alright, well thank you Micheal.
  133. >> Sure.
  134. >> I will see you next time we're in the same place, at the same time.