1, the return is a cousin of Darwin discovered, that mankind in general has an average height. The child born of tall are high, short children born short. But most of them to their children long average height, that is particularly high, their children will be high, but often shorter than their own. Especially short people they will be short some of the children born, but will be higher than their parents, they have this tendency. This is called cousin pipe return.
2, a linear regression refers to one independent variable corresponds to a unary function dependent variable. When the point of the plane coordinate system some hash, we often want to find a straight line to fit their regression line. We are in charge of this line is called linear regression
This function is assumed 
we ask is this function 
and 
determine this value we determine the value of the linear regression function
3, the cost function is equal to
It is clear that if 
the value can make 
the minimum that is a linear regression function we want.
4, gradient descent
Is seeking partial differential cost function for two variables, then this partial differential value is multiplied by a variable so that the current value of the derivative - * variable amount of learning. Is the minimum value of the gradient descent method desired. At this point parameters
We need two values is a linear regression equation. About learning variables do not think it is very mysterious to get the coordinates to study it only in order to meet the derivation of the current variable slope change direction and magnitude of movement in the coordinate system and nothing more
If it is not about the divergent - but the worst case is seeking + peak at this time of fitting the maximum cost function
Another method Why not a little calculus twice in the derivative's extreme value