CS 4700: Foundations of Artificial Intelligence
description
Transcript of CS 4700: Foundations of Artificial Intelligence
![Page 1: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/1.jpg)
1
CS 4700:Foundations of Artificial Intelligence
Prof. Bart [email protected]
Machine Learning:The Theory of Learning
R&N 18.5
![Page 2: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/2.jpg)
2
Machine Learning Theory
Central question in machine learning:
How can we be sure that the hypothesis produced by a learning alg. will produce the correct answer on previously unseen examples?
Question in some sense too vague… Leads to the general problem of inductive inference: How can we generalize
from data?
Major advance: Computational Learning Theory Valiant 1984; Turing Award 2010.
![Page 3: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/3.jpg)
3
![Page 4: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/4.jpg)
4
![Page 5: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/5.jpg)
5
![Page 6: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/6.jpg)
6
Probabilities to the rescue: certain “bad” events (e.g.learner learns the wrong hypothesis)become exponentially rare with enough data.
![Page 7: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/7.jpg)
7
![Page 8: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/8.jpg)
8
(or higher.)
![Page 9: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/9.jpg)
9
S = # of “heads”
![Page 10: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/10.jpg)
10
Wow!
S = # of “heads”
So, when flipping a fair coin 10,000 times, you’ll“never, ever” see more than 5,500 heads. And, no oneever will… I.e., with enough data, we can be verycertain that if coin is fair, we won’t see a “large” deviationin terms of #heads vs. #tails.
![Page 11: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/11.jpg)
11
![Page 12: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/12.jpg)
12
![Page 13: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/13.jpg)
13
![Page 14: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/14.jpg)
14
Also, makes interesting learning algorithms possible!
![Page 15: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/15.jpg)
15
![Page 16: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/16.jpg)
16
Two issues: (1) (Re: High prob.) May get a “bad” sequence oftraining examples (many “relatively short men). Smallrisk but still small risk of getting wrong hypothesis.(2) We only use a small set of all examples (otherwise notreal learning). So, we may not get hypothesis exactly correct.
Finally, want efficient learning --- polytime!!
![Page 17: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/17.jpg)
17
Valiant’s genius was to focus in on the “simplest” modelsthat still captured all the key aspects we want in a “learningmachine.” PAC role has been profound even though mainlyto shine a “theoretical” light.
![Page 18: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/18.jpg)
18
![Page 19: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/19.jpg)
19
![Page 20: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/20.jpg)
20
For our learning algorithm, we simply use a method thatkeeps the hypothesis consistent with all examples seenso far. Can start out with an hypothesis that says “No”to all examples. Then when first positive example comesin, minimally modify hypothesis to make it consistent with thatexample. Proceed doing that for every new pos example.
![Page 21: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/21.jpg)
21
So, h_b could mistakenly be learned!
Note: we want to make it likely that all consistenthypotheses are approximately correct. So, no “bad” consistenthypothesis occur at all.
![Page 22: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/22.jpg)
22
² and ± are assumed given(set as desired)Keep size hypothesis class H down.Another showing of Ockham’s razor!
![Page 23: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/23.jpg)
23
So, in this setting the requirements on our learningalgorithm are quite minimal. We do also want poly time though.
![Page 24: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/24.jpg)
24
Aside: Shannon already noted that the vast majorityof Boolean functions on N letters look “random” andcannot be compressed. (No structure!)
![Page 25: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/25.jpg)
25
![Page 26: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/26.jpg)
26
![Page 27: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/27.jpg)
27
![Page 28: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/28.jpg)
28
![Page 29: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/29.jpg)
29
![Page 30: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/30.jpg)
30
![Page 31: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/31.jpg)
31
![Page 32: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/32.jpg)
32
Still need polytime learning alg. to generatea consistent decision list with m examples.
![Page 33: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/33.jpg)
33
![Page 34: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/34.jpg)
34
![Page 35: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/35.jpg)
35
![Page 36: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/36.jpg)
36
![Page 37: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/37.jpg)
37
![Page 38: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/38.jpg)
38
![Page 39: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/39.jpg)
39
![Page 40: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/40.jpg)
40
Some more PAC learnability examples.
![Page 41: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/41.jpg)
41
![Page 42: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/42.jpg)
42
![Page 43: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/43.jpg)
43
![Page 44: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/44.jpg)
44
![Page 45: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/45.jpg)
45
![Page 46: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/46.jpg)
46
![Page 47: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/47.jpg)
47
![Page 48: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/48.jpg)
48
![Page 49: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/49.jpg)
49
![Page 50: CS 4700: Foundations of Artificial Intelligence](https://reader036.fdocuments.in/reader036/viewer/2022070503/56816267550346895dd2d21d/html5/thumbnails/50.jpg)
50