When it comes to Linear Regression Logistic regression or other key words, it will involve a concept called Likelihood Function and Maximum Likelihood Estimation and so on. Chinese translation is called the "likelihood estimates", according to my own understanding, that this translation is not so appropriate, just as "robust" as.
The following is my understanding and interpretation of this concept, if you can promote understanding, is my honor, if there is an error, please also spending:
Mathematics compulsory education, examination paper above topics, the condition is directly to you, for example, let you solve an equation: y = 2x + 1 and y = 0 or solution y = ax + b, but the answer which allows you with the parameters a and b parameters. We can call them model .
Into the university, we want to progress, to be more fit reality. In practical applications, we can not come in handy to get a perfect model, or a model with all the exact parameters. This guide book has a lot of feelings in the world is similar, but what this law did not take your girlfriend whole instill obedience (provided you have).
Then we should do to get a good model? You can make use of the hands of existing resources, experience or data, and then substituting them into your good model assumptions in advance and then get a more satisfactory and ideal parameters. Note that this hypothetical model is based on experience. A innocent girl, and a violent temper girl, model assumed in advance are poles other.
In probability theory, this model is the probability density data is the data (x, y), parameter is the coefficient $ \ theta $. This is Linklihood Function, parameter is unknown quantities, x and y are known quantities, but also a lot of x, y.
Maximum means that the probability density we finally obtained, is closest to the real situation. Of course, most real situation, who do not know, probably not a mathematical field, but the field of philosophy instead.