›
›

Bài giảng Trí tuệ nhân tạo - Chapter 11: Machine learning

Download
Machine Learning  
Chapter 11  
Machine Learning  
• What is learning?  
2
Machine Learning  
• What is learning?  
• “That is what learning is. You suddenly understand  
something you've understood all your life, but in a  
new way.”  
(Doris Lessing – 2007 Nobel Prize in Literature)  
3
Machine Learning  
• How to construct programs that automatically  
improve with experience.  
4
Machine Learning  
• How to construct programs that automatically  
improve with experience.  
• Learning problem:  
– Task T  
– Performance measure P  
– Training experience E  
5
Machine Learning  
• Chess game:  
– Task T: playing chess games  
– Performance measure P: percent of games won against  
opponents  
– Training experience E: playing practice games againts itself  
6
Machine Learning  
• Handwriting recognition:  
– Task T: recognizing and classifying handwritten words  
– Performance measure P: percent of words correctly  
classified  
– Training experience E: handwritten words with given  
classifications  
7
Designing a Learning System  
• Choosing the training experience:  
– Direct or indirect feedback  
– Degree of learner's control  
– Representative distribution of examples  
8
Designing a Learning System  
• Choosing the target function:  
– Type of knowledge to be learned  
– Function approximation  
9
Designing a Learning System  
• Choosing a representation for the target function:  
– Expressive representation for a close function approximation  
– Simple representation for simple training data and learning  
algorithms  
10  
Designing a Learning System  
• Choosing a function approximation algorithm  
(learning algorithm)  
11  
Designing a Learning System  
• Chess game:  
– Task T: playing chess games  
– Performance measure P: percent of games won against  
opponents  
– Training experience E: playing practice games againts itself  
– Target function: V: Board → R  
12  
Designing a Learning System  
• Chess game:  
– Target function representation:  
V^(b) = w0 + w1x1 + w2x2 + w3x3 + w4x4 + w5x5 + w6x6  
x1: the number of black pieces on the board  
x2: the number of red pieces on the board  
x3: the number of black kings on the board  
x4: the number of red kings on the board  
x5: the number of black pieces threatened by red  
x6: the number of red pieces threatened by black  
13  
Designing a Learning System  
• Chess game:  
– Function approximation algorithm:  
(<x1 = 3, x2 = 0, x3 = 1, x4 = 0, x5 = 0, x6 = 0>, 100)  
x1: the number of black pieces on the board  
x2: the number of red pieces on the board  
x3: the number of black kings on the board  
x4: the number of red kings on the board  
x5: the number of black pieces threatened by red  
x6: the number of red pieces threatened by black  
14  
Designing a Learning System  
• What is learning?  
15  
Designing a Learning System  
• Learning is an (endless) generalization or induction  
process.  
16  
Designing a Learning System  
Experiment  
Generator  
New problem  
(initial board)  
Hypothesis  
(V^)  
Performance  
System  
Generalizer  
Solution trace  
(game history)  
Training examples  
{(b1, V1), (b2, V2), ...}  
Critic  
17  
Issues in Machine Learning  
• What learning algorithms to be used?  
• How much training data is sufficient?  
• When and how prior knowledge can guide the learning process?  
• What is the best strategy for choosing a next training experience?  
• What is the best way to reduce the learning task to one or more  
function approximation problems?  
• How can the learner automatically alter its representation to  
improve its learning ability?  
18  
Example  
Experience  
Example  
Sky  
AirTemp Humidity  
Wind  
Water  
Forecast EnjoySport  
1
2
Sunny Warm  
Sunny Warm  
Normal  
High  
Strong Warm  
Strong Warm  
Same  
Same  
Yes  
Yes  
3
4
Rainy Cold  
Sunny Warm  
High  
High  
Strong Warm  
Strong Cool  
Change  
Change  
No  
Yes  
Low  
Weak  
Prediction  
5
6
7
Rainy Cold  
High  
Strong Warm  
Strong Warm  
Strong Cool  
Change  
Same  
?
?
?
Sunny Warm  
Sunny Warm  
Normal  
Low  
Same  
19  
Example  
• Learning problem:  
– Task T: classifying days on which my friend enjoys water sport  
– Performance measure P: percent of days correctly classified  
– Training experience E: days with given attributes and classifications  
20  
Tải về để xem bản đầy đủ
pdf 55 trang myanh 21620 Free
Download
Bạn đang xem 20 trang mẫu của tài liệu "Bài giảng Trí tuệ nhân tạo - Chapter 11: Machine learning", để tải tài liệu gốc về máy hãy click vào nút Download ở trên

File đính kèm:

  • pdfbai_giang_tri_tue_nhan_tao_chapter_11_machine_learning.pdf