Well posed Machine Learning

 

Well posed Machine Learning

Simply to say precisely it is nothing but  learning by experience.

An agent or a Person solves a problem or Task T, Performance P, gain some experience E.

By doing this task T with performance P this agent will give some experience is denoted by E.

If P is the person measured at T it can improve E, with the help of Performance it will increase the experience.

 

If the performance is good, he will try to improve more good and become best.

If the performance is not good in that case he will understand law pols he will try to improve the experience he had gain through that, used some experience he adds some more new methodology to improve more that is well posed learning problem.

Example:

• Playing Checkers Problems
• Hand Written Recognition Problem
• Robot Driving Learning Problem

Let us see in all these there scenario what is the task T, what is the Performance P and what is the experience E that the learner will get some particular task.

Here if you can see that here the problem

Task 1 : what the user support to do

Task 2: how the user Performance should be their what is the user experience were is going to out of that.

Problem	Task (T)	Performance (P )	Experience( E)
Playing Checkers Learning	Playing against opponents to win the game If you are writing will be an exam,  your task is finish  all the questions in the given time in checkers need to win over the opponents	Make perfect moves in order to win the game So that your Performance is good If you are not making perfect moves so that your moves get wasted Example: Candy Crush	It plays itself to improve
Handwriting Recognition Learning	Classifying the images and Text	Better Classification	A database of homework test that means it have to some inbuilt database so that observing the observation of the data from the database it will learn itself on it will perform
Robot Driving Learning Problem	Drive the car in a four lane highway	Source to destination the average distance travelled should be longer and safer	Image vehicle on road
Face Recognition Problem	Redirecting different types of data	Measurable to predict maximum type of faces	Training machine with maximum amount of dataset of different face images
To better filter emails as spam or not	Classifying emails as spam or not	Measures the fraction of email accordingly classified as a spam or not	Observing you label email as spam or not
Fruit prediction problem	Forecasting different fruits for recognition	Measurable to predict maximum variety of fruits	Balancing the machine with a large datasets of fruits image
Automatic translation of documents	Translating one type language used in a document to other languages	Measurable to convert one language to another efficiency	Training machine with a large amount of dataset of different types of languages