Note for video Machine Learning and Data Mining——Theory of Generalization

Here is the note for lecture six.

Today I just note something about the number of hypothesis in a set with a break point can be replaced by a polynomial.

What is break point?

If the number of uncertain is a, and choice for a uncertain number is b, the number of different hypothesis we can get is
a^b. But actually, there are some hypothesis that cannot be built up, and break point exists. And if there are x uncertain 
number and hypothesis that cannot be built up, we call the hypothesis set's break point is x.

Now give a combinatorial quantity B(n,k). There are n points, and k is a break point. Then how many different patterns can 
you get? And can you lead these patterns to a polynomial?
B(n.k) means number of different patterns. There will be a inequlity
                                    B(n,k) < = B(n-1,k)+B(n-1,k-1)
According to the recursion, we can get a theorem by induction. So what does the theorem?
 
And it is easy to know from the theorem that B(n,k) can be replaced by a polynomial.