Re: [agi] general patterns induction

[Shane Legg]

Pablo,

If you are interested in Solomonoff induction and don't want to
spend all the money to buy the book just yet then you might be
interested in a paper I wrote a few years back that starts from
basic computation theory and proves all the key results.
It about 25 pages of mathematics in PDF format, I can email it
to you if you like.

The key points are:

Solomonoff induction will learn anything that is computationally
expressible (i.e. anything useful) with an error rate that falls
to zero faster than 1/n where n is the number of bits of input
data that the system has been given.

Solomonoff induction is not computable and it is difficult to
approximate well.

Most learning methods in statistics and machine learning can in
some sense be proven to be computable approximations to Solomonoff
induction.


So, in short, it's an interesting theoretical model that's amazingly
powerful but it's not something you can directly use in practice.
The prize for an amazingly powerful and practical system is still
very much up for grabs :)

If you're serious about data compression also check out a book
called Text Compression by Bell, Cleary and Witten.

Cheers
Shane


-------
To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]