Fundamental Compressionist Philosophy.

[Brendan Macmillan]

Detlef,

> It may be boring, but again: We need not compress the data.
> We must compress **the theory**!
Is there a distinction?

> When a significantly smaller theory can be found which reproduces the
> initial set, we have 
> (1) an abstract representation of the law which generated the data
> (2) a means of predicting results for samples which until know had
> never been observed!!
Yes, exactly right.  The more compact the representation, the better it
generalizes to new, unseen cases.


Brendan