• If you are citizen of an European Union member nation, you may not use this service unless you are at least 16 years old.

• You already know Dokkio is an AI-powered assistant to organize & manage your digital files & messages. Very soon, Dokkio will support Outlook as well as One Drive. Check it out today!

View

# Scribe Note 7-1

View current version     Page history
Saved by
on December 9, 2011 at 3:39:33 pm

Q6

a)We have defined the information rate distortion function as

where the minimization is over all conditional distributions    for which the joint distribution   satisfies the expected distortion constraint. This is a standard minimization problem of a convex function over the convex set of all  satisfying  for all  and

We can use the method of Lagrange multipliers to find the solution.  We set up the functional

Differentiating with respect to  , setting , we obtain

Since  , we must have   or

for all . We can combine these  equations with the equation defining the distortion and calculate λ and the   unknowns . We can  find the optimum conditional distribution.

The above analysis is valid if    is unconstrained. The inequality condition  is covered by the Kuhn–Tucker conditions, which reduce to

Substituting the value of the derivative, we obtain the conditions for the minimum as

This characterization will enable us to check if a given  is a solution to the minimization problem. However, it is not easy to solve for the optimum output distribution from these equations. In the next section we provide an iterative algorithm for computing the rate distortion function. This algorithm is a special case of a general algorithm for finding the minimum relative entropy distance between two convex sets of probability densities.

b)

Consider the following problem: Given two convex sets A and B in Rn as shown in following figure, we would like to find the minimum distance between them:

where d(a, b) is the Euclidean distance between a and b. An intuitively obvious algorithm to do this would be to take any point x ∈ A, and find the y ∈ B that is closest to it. Then fix this y and find the closest point in A. Repeating this process, it is clear that the distance decreases at each stage. Does it converge to the minimum distance between the two sets? Csiszar and Tusnady have shown that if the sets are convex and if the distance satisfies certain conditions, this alternating minimization algorithm will indeed converge to the minimum. In particular, if the sets are sets of probability distributions and the distance measure is the relative entropy, the algorithm does converge to the minimum relative entropy between the two sets of distributions.

To apply this algorithm to rate distortion, we have to rewrite the rate distortion function as a minimum of the relative entropy between two sets.