classes of ml

what is machine learning?

: designing algorithms for inferring what is unknowns from knowns.

 

algorithm

inference

 

applications :

– spam detection

– handwriting detection

– in google streetview : blurring out people’s faces (face detection) / license plates

– speech recognition

– netflix : preference, recommendation

– navigation for robot : integrating all sensor, video data, etc

– climate modelling : weather pattern prediction

 

 

 

_Classes of ML problems

 

supervised learning

: canonical

given (x1, y1) … (xn, yn)

xi : data points

yi : target value

Find a function f(x) = y

– classification : yi belong to finite set

– regression : yi in real value

E4DD8957-B8DE-4364-8E31-B4DC2F417D5F.png

 

 

 

classiification :

이 때 x는 벡터일 수도 있다.  xi = (xi_1, xi_2)

781D3393-DF6B-425E-AC5C-45B04D3A7EEE.png

 

이때 새로운 xn이 주어졌을 때 yn은 0일 것인지 1일 것인지 판단하는 것.

 

regression :

E9F5D56D-762A-449F-92C4-B69185777D2D.png

 

새로운 any given x에 대해서 y의 값을 예측하는 function을 찾아내는 것.

 

 

 

unsupervised learning

 

given (x1, … , xn) (각 x는 여러 dimension이 있을 수 있음.)

57EF4A5E-D530-4B43-A130-9808E6E040A8.png

 

find patterns in the data.

– clustering

– density estimation

– dimenstionality reduction

 

_clustering

AAD6C1C5-5AE0-46AB-817F-81A74C668C48.png

: 이렇게 데이터를 어떤 특성에 따라 나누는 cluster를 찾아내는 것.

 

_density estimation

assuming these data xi come from some probability distribution,

and u want to estimate the density.

 

 

 

_dimensionaliity

FCBCF606-8920-41E0-AF57-DFACE0B2B047.png

find some lower dimensional space that u can represent this data lies on.

5F698AB9-470F-4CA4-90FA-9093CEE49014

 

each point will be projected on to lower dimensional dignosis

 

 

 

* while preserving the structure of the data

(단순한 projection으로는 structure 정보가 날라간다.)

 

 

 

 

Variations

 

_ semi-supervised learning

 

given

(x1, y1) … (xk, yk), xk+1, ….. xn

 

predict the labels for the remaning points

(some of values are missing (ex. Netflix problem : 중간중간 빈 데이터들이 있다.))

 

혹은, 다음과 같은 문제.

9A0D8DB3-8137-4714-8C7B-67810B988F8B

 

 

 

active learning

: ask for information (labels)

 

 

decision theory

: measuring performance (how many u got wrong)

 

 

 

 

 

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s