For this class, the lecturer talk about the sample definition of machine learning.
This explanation is called EPT.
E is the experience, which means the experience of lots of games of checkers against itself.
P is the performance measure, which means the fraction of games it wins against a certain set of human players.
T is the task, which means the task of playing checkers.
In this class, there are four parts contained into the machine learning topic: supervised learning, unsupervised learning, learning theory as well as reinforcement learning.
The supervised learning is to supervise the algorithm with giving the algorithm a quote of right answer for a number of houses.
1)regression -->continuous value
such as the relation between the house area and the value of houses.
2)classification -->discreet (always as 1 or 0)
such as the relationship between the malignant or not of tumor cancer and the tumor size or the age of the patient.
The unsupervised learning is the application of clustering algorithm. As the lecturer introduced in the class, the cocktail party problem is an example of this algorithm.
The typical application of unsupervised learning is ICA algorithm and the least square algorithm.(To be honest, I cannot believe that I still need to learn something about least square, I feel really, just enough:) )
About the reinforcement learning, the reward faction is the method to achieve the machine learning, we can also understand it as the way to impressive, develop and improve the memory of experience.
Really useful class, hope I can insist to learn.