Re: Entropy and Quantity of Information

From: Joe Thurbon <usenet_at_thurbon.com>
Date: Sat, 12 Jan 2008 00:38:50 GMT
Message-ID: <u_Thj.2146$421.324_at_news-server.bigpond.net.au>


David Cressey wrote:
> Instead of hijacking another topic, I'll start this topic.
>
> David BL suggested that a way to quantify information is the number of bits
> needed to encode, allowing for compression.

I think the term being looked for is

http://en.wikipedia.org/wiki/Kolmogorov_complexity

There's been a _lot_ of work done in this area.

> I said I preferred entropy as
> the measure of information, and suggested that the two measures might in
> some way be equivalent. Someone else recalled the concept of entropy from
> the study of statistical mechanics and thermodynamics in physics.

It also gets a good work out in machine learning. Many of the classical algorithms (decision trees in their various forms) have information-gain (which is defined in terms of entropy) at their heart. (The precise definition is closely related to the number of bits required for representation)

I vaguely recall that some of the more theoretical machine learning results (like learnability and optimality results) rely on a notion entropy.

[... snip interesting perspective ...]

> All of this goes back to the 1960s, and some of it to the 1940s. Is entropy
> still widely used in information science?

Yes.

> Is it relevant to database theory?

Good question.

Cheers,
Joe Received on Sat Jan 12 2008 - 01:38:50 CET

Original text of this message