This preview shows pages 1–3. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Harvard SEAS ES250 – Information Theory Differential Entropy and Maximum Entropy * 1 Differential Entropy 1.1 Definitions Definition The differential entropy h ( X ) of a continuous random variable X with density f ( x ) is defined as h ( X ) =- integraldisplay S f ( x ) log f ( x ) dx, where S is the support set of the random variable. 1.2 AEP for Continuous Random Variables Theorem Let X 1 ,X 2 ,...,X n be a sequence of random variables drawn i.i.d. to the density f ( x ). Then- 1 n log f ( X 1 ,X 2 ,...,X n ) → E [- log f ( X )] = h ( X ) in probability . Definition For epsilon1 > 0 and any n , we define the typical set A ( n ) epsilon1 with respect to f ( x ) as follows: A ( n ) epsilon1 = braceleftbigg ( x 1 ,x 2 ,...,x n ) ∈ S n : vextendsingle vextendsingle vextendsingle vextendsingle- 1 n log f ( x 1 ,x 2 ,...,x n )- h ( X ) vextendsingle vextendsingle vextendsingle vextendsingle ≤ epsilon1 bracerightbigg , where f ( x 1 ,x 2 ,...,x n ) = producttext n i =1 f ( x i ). Definition The volume Vol( A ) of a set A ⊂ R n is defined as Vol( A ) = integraldisplay A dx 1 dx 2 ··· dx n . Theorem The typical set A ( n ) epsilon1 has the following properties: 1. Pr( A ( n ) epsilon1 ) > 1- epsilon1 for n sufficiently large. 2. Vol( A ( n ) epsilon1 ) ≤ 2 n ( h ( X )+ epsilon1 ) for all n . 3. Vol( A ( n ) epsilon1 ) ≥ (1- epsilon1 )2 n ( h ( X )- epsilon1 ) for n sufficiently large. Theorem The set A ( n ) epsilon1 is the smallest volume set with probability ≥ 1- epsilon1 , to first order in the exponent. * Based on Cover & Thomas, Chapter 8 and 12 1 Harvard SEAS ES250 – Information Theory 1.3 Relation of Differential Entropy to Discrete Entropy Assume the density is continuous. Then, by mean value theorem, there existsAssume the density is continuous....
View Full Document
This note was uploaded on 12/01/2010 for the course ADLAC 1023 at Stanford.