Talk:Rate–distortion theory

Latest comment: 2 years ago by Nathanielvirgo in topic D not defined

The next thing to discuss is decorrelating the images, for example using the K-L transform, and you can then treat them as independent/memoryless...

I wish the page was a little more coherent and organized - for example D* is presented at the top of the page but not explained until much later.

[speedplane] I really like the example given on the rate distortion function of a memoryless guassian source... but a reference to the is necessary. Just showing the equation explains little about the intuition.

Homework problem

edit

I'm not sure the equations for H(Y) and H(Y|X) are correct. I don't understand information theory well enough to correct it, but I'd appreciate it if someone could verify. In particular, from what I can understand  , taking the integrand evaluated at 0 to be 0. Those are base 2 logs, naturally. This misunderstanding is making doing my homework difficult:-) --Calmofthestorm7 (talk) 17:19, 26 December 2008 (UTC)Reply

I believe your formula is correct, and that there's a minus sign missing in the article (it's usually written with the -log(p) instead of log(1/p), which is the same thing). The log base determines the units; 2 for bits, e for nats. But why not just look in the sources and correct the article, instead of trying to rely on wikipedia, which is never a good idea? Dicklyon (talk) 00:03, 27 December 2008 (UTC)Reply
I went ahead and added a couple of minus signs, but it will take some time studying sources like these to get the article verified and referenced. Dicklyon (talk) 00:20, 27 December 2008 (UTC)Reply

See also: Sphere packing

edit

Why is sphere packing in the "See also" list for this article? I don't see the connection. --A3 nm (talk) 09:33, 21 February 2014 (UTC)Reply

D not defined

edit

The capital 'D's in this article appear without being given a definition, which makes the article impossible to follow. Can a definition be added? Nathaniel Virgo (talk) 02:43, 24 March 2022 (UTC)Reply

Come to think of it, P_X and P_Y are not defined either. Can we assume that P_Y is fixed (since the optimisation appears to be over Q) and that P_X(x) = sum_y P_Y(y)Q(y|x)? And D_Q is \sum_{x,y} P_Y(y)Q(y|x)d(x,y)? Nathaniel Virgo (talk) 02:48, 24 March 2022 (UTC)Reply