7. Definition( Machine Learning / ๊ธฐ๊ณํ์ต )
A set of methods that can automatically detect patters in data, and then use
the uncovered patterns to predict patterns to predict future data, or to
perform other kinds of decision making under uncertainty
"Machine Learning-A Probabilistic Perspective"
Kevin P. Murphy
9. A feature vector is an ๐-dimensional vector of numerical features
that represent some object.
For example ,
a document , ๐ฅ๐ : ๋ฌธ์ ์์์์ ๐ ๋ฒ์งธ ๋จ์ด
I love you. ๐ฉ = ๐ผ, ๐๐๐ฃ๐, ๐ฆ๐๐ข
10. ๐ฉ ๐ฆ
๐mapping
๋ถ์ฐ์ ๊ฐ ๋๋
์ฐ์ ๊ฐ
E-mail (Words) Spam or Not(๋ถ)
Web Site (Words) Sports or Science or News(๋ถ)
ํน์ฑ ๋ฒกํฐ
๊ฝ (๊ฝ์ ์๊น์) Line-flower or Mass-flower(๋ถ)
์์ด์ ํค ์๋ฒ์ง์์ ํค(์ฐ)
๋ฐฉ์ ๊ฐ์, ๋ฐฉ์ ๋์ด ์ง ๊ฐ(์ฐ)
(visit, money, buy, girl, Viagra)
For example,
Spam mail
๐
11. Goal of Supervised Learning (predictive learning)
To learn a mapping(function) from input ๐ฉ to output ๐ฆ, given a labeled set of
input-output pairs ๐ท = ๐ฉ๐, ๐ฆ๐ ๐=1
๐
, where ๐ท is called the training set, and ๐
is the number of training examples.
- ๐ฉ๐ : ๐ท-dimensional vector of numbers โ feature vector
- ๐ฆ๐ : response variable โ categories or real-values
๐ฆ๐ is categorical, the problem is classification,
๐ฆ๐ is real-valued, the problem is regression.
12. What is natural grouping among these objects?
Simpson's Family School Employees Females males
13.
14. Goal of Unsupervised Learning (descriptive learning)
To find "interesting pattern" in the data, given ๐ท = ๐ฉ๐ ๐=1
๐
, where ๐ท is called
the training set, and ๐ is the number of training examples.
- ๐ฉ๐ : ๐ท-dimensional vector of numbers โ feature vector
It is known knowledge discovery.
15. Goal of Reinforcement Learning
To learn how to act or behave when given occasional reward or punishment
sinnals.
17. ๐ฉ ๐ฆ
๐mapping
๋ถ์ฐ์ ๊ฐ ๋๋
์ฐ์ ๊ฐ
E-mail (Words) Spam or Not(๋ถ)
Web Site (Words) Sports or Science or News(๋ถ)
ํน์ฑ ๋ฒกํฐ
๊ฝ (๊ฝ์ ์๊น์) Line-flower or Mass-flower(๋ถ)
18. Goal of Classification
To learn a mapping(function) from input ๐ฉ to output ๐ฆ, where ๐ฆ โ {1, โฆ , ๐ถ},
with ๐ถ being the number of classes.
function approximation
Assume that ๐ฆ = ๐(๐ฉ) for unknown function โ, the goal of learning is to
estimate function ๐ given a labeled training set, and then to make predictions
using ๐ฆโ = ๐โ(๐ฉ). Then we can make predictions on novel input.
way to formalize the problem
19. Compute!! our "best guess" using
๐ฆโ = ๐โ ๐ฉ = ๐๐๐ max
๐=1โฆ๐ถ
๐(๐ฆ = ๐|๐ฉ, ๐ท)
This corresponds to the most probable class label.
It is known as a MAP estimate (Maximum a Posterior).
Ex) mail spam filtering
๐ ๐ฆ = 1 ๐ฉ, ๐ท)
๐ ๐ฆ = 2 ๐ฉ, ๐ท)
...
๐ ๐ฆ = ๐ถ ๐ฉ, ๐ท)
20. Goal of Regression
To learn a mapping(function) from input ๐ฉ to output ๐ฆ, where ๐ฆ is continuous.
๐ฆ = ๐1 + ๐2 ๐ฅ ๐ฆ = ๐1 + ๐2 ๐ฅ + ๐3 ๐ฅ2
22. What is natural grouping among these objects?
Simpson's Family School Employees Females males
23.
24. Goal of Clustering
To estimate the distribution over the number of cluster, ๐ ๐พ ๐ท ; this tells us if
there are subpopulations within the data
To estimate which cluster each point belong to.
25. ๐พโ
= ๐๐๐ max
๐พ
๐(๐พ|๐ท)
We often approximate the distribution
We can infer which cluster each data point belongs to by computing
๐ง๐
โ
= ๐๐๐ max
๐
๐(๐ง๐ = ๐|๐ฉ๐, ๐ท)
๐ ๐ง๐ = 1 ๐ฉ, ๐ท)
๐ ๐ง๐ = 2 ๐ฉ, ๐ท)
...
๐ ๐ง๐ = ๐ถ ๐ฉ, ๐ท)
26.
27. Goal of Dimensional Reduction
To reduce the dimensionality by projecting the data to a lower dimensional
subspace which captures the "essence" of the data.
Motivation : Although the data may appear high dimensional, there are
only be a small number of degrees of variability, corresponding
to latent factors.
latent factor : which describe most of the variability
28. ๏ฌ "Machine Learning-A Probabilistic Perspective" Kevin P. Murphy
๏ฌ http://ko.wikipedia.org/wiki/๊ธฐ๊ณ_ํ์ต