On Sat, Jan 29, 2005 at 10:43:37AM -0600, Jan Depner wrote: > On Sat, 2005-01-29 at 06:44, Fons Adriaensen wrote:
> > That's a classic one. For large N, the output will approach > > some form of filtered Gaussian noise. What makes this fractal > > noise superior as a source of randomness in a resample algo ? > > Because it more closely resembles nature. Take a look at this page: > > http://swiss.csail.mit.edu/~rauch/islands/ Nice pictures. But: - Not everything in nature is fractal. Some things (even very complex ones) are very clearly not. - We are not trying to mimic nature, just to avoid a nasty comb filtering effect. - If I gave you two series of samples, one generated with the fractal method, and one generated by sending white Gaussian noise through a suitable filter, you would have no way to tell which is which. And that means there is nothing special about the fractal noise, apart from the fact that is was generated by an interesting algo :-) -- FA
