MachineLearningFoundation@NTU(HsuanTien.LIN)

1.what is machine learning ?which problem we will use machine learning?
if you want to use machine learning to solve problem,you should can offer large number of data, and the problem should has the probability to solve, and you may van predict the result. the problem will be much difficult to be solve by the tranditional compute method.

2.To build a movie recommend system ,what can we do with machine learning?
A Possible ML Solution:
when we let the app/system user to upload the movie or we upload the movie by ourselves ,we should suggest them to upload the movie with tags(so call :factors).then we can learn from the data(movie factor) when a user is using it, then we can know people like which kind of movie and we can also recommend the same kind of movie to the users when they use it next time.

3. The use of the Machine Learning in our real life:

4.Normalization:
https://blog.csdn.net/leiting_imecas/article/details/54986045

5.Supervised Learning:Regression and Classification
(a) Regression - Given a picture of a person, we have to predict their age on the basis of the given picture
(b) Classification - Given a patient with a tumor, we have to predict whether the tumor is malignant or benign.

Unsupervised Learning:‘Clustering’ and ‘Non-clustering’
Clustering: Take a collection of 1,000,000 different genes, and find a way to automatically group these genes into groups that are somehow similar or related by different variables, such as lifespan, location, roles, and so on.
Non-clustering: The “Cocktail Party Algorithm”, allows you to find structure in a chaotic environment. (i.e. identifying individual voices and music from a mesh of sounds at a cocktail party).

6.there has a time that :

7.Matrix && Vector:
An m x n matrix multiplied by an n x 1 vector results in an m x 1 vector.

perceptron:

why when we use the algorithm, we let the divide border be vertical with the vector “w”?
may. because when a linear be vertical with vector “w”,(the linear represent vector"x" in the plane which plot on the border line )the “(w^T) *x” will be zero which mean linear can be as a divided border.when other x sample do the same things with “w”,it will has a result as “bigger than 0” or “small than 0”,the we can differentiate the different class ,and then we can get the result.may there will has a sample when it match the “x” cause a result what we do not want to see ,we can use w(t+1)<—w(t)+y(n(t))x(n(t)) to get a more accurate “w”, and we use this repeat method to find the best “w” (which is the best hypothesis, the “g”).

here we can see that we use

here we can see when there is a “w(f)^T” which can match every example ,which means we can calssfity then different class with no mistake.
“w(f)^T” and “w(t)” will be more and more closer when we try to correct the mistake.

here

will return the result of cos(theta),which measure the close of the “w(f)^T” and “w(T)” .and we can see it is positive reaction with the “T(which represent how many time have we interate)”, so we can come the result that the “w(T)” is become much more closer with “w(f)^T”,which means we are closer with function “f”.

9.Multiclass Classification:

different type of output space “y”:

10.the data(sample) should be enough big, if not it will cause E(in)small but E(out) big.the Hyhpothesis Set you make should be enough big, if not the particular “h” will cause only one choice of the “h” toilet you choose ,so that you can not choose a good “g”.

10.