Machine Learning 学习日记

Few days before i did some works for my friend in lib about meat classfication problem.A model we used is SVM . From the result ,let's talk about the bias and variance in Machine Learning.

Pic.1(from Andrew Ng)

So what is high bias situation? In pic 1, I draw a line(maybe a curve) to justify two different symbols  .You may find out that this line could not distinguish most of data , expect the right one . In ML only the mid one is the best case we want to see . The left one corresponding to under-fitting. The machine learns few information from the data set. And the right on learning too much ,we call it over-fitting . One reason we do not want model like the right one is that this might cause model stubborn . High bias situation corresponds to under-fitting ,high variance corresponds to over-fitting. A trade-off between bias and  variance cause the mid one: "just right",a great balance.

In deep learning ,especially with huge data , get high accuracy is really easy ,so long as your're regularizing and your model makes sense.