From: Perry E.Metzger (email@example.com)
Date: Wed Dec 03 2003 - 19:25:00 MST
"Eliezer S. Yudkowsky" <firstname.lastname@example.org> writes:
>> And noise can make a weak signal detectable ("stochastic resonance"):
> One of these papers describes how adding a noise term to a weak input
> signal to a cricket *neuron* resulted in more information in the spike
> train emerging *from the cell*; it doesn't mean that adding noise to a
> weak signal actually adds information to it!
It is not a question of adding information. It is a question of being
able to do efficient signal processing to remove the desired
information. It has been known that noise injection can improve
certain kinds of signal processing for many years.
> Perhaps the cricket cells evolved to work in the
> presence of noise, or in the presence of sound at a particular
> threshold value, but that does not mean the noise "boosts processing
> efficiency"; an ab initio algorithm could extract more data from the
> signal without the noise.
Not true. There are known conditions in which, as I said, injecting
noise makes it easier to extract signal. It doesn't add information,
of course -- this is just a practical signal processing expedient.
Why does this bother you so much?
> Noise is not magic, and for noise to result in any fundamental
> algorithmic improvement would, I maintain, violate the second law of
I can't possibly see why. There are hundreds of problems, for example,
in which the best probabilistic time complexity is known to be far
lower than the best non-probabilistic time complexity. Big deal.
Get used to randomness being a useful tool -- as Chaitin more or less
proved (the more or less depending on how you choose to interpret the
result), math itself is in a certain sense random.
-- Perry E. Metzger email@example.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT