I did a quick search but found mostly various speculations that the brain "seems to do quick FFT", and various _models of the brain_ using fourier transforms in the pattern recognition. I see the clear association there, but it's on the same level as nature. I am trying to understand this deeper and how it connects to a logical description that can be formalised.

 

The fourier transform has interesting properties, for example the gaussian distribution (beeing central in probability theory) is self-dual and transforms to itself. The fourier transform of a gaussian distribution is also gaussian.

 

However, the gaussian is not the only self-dual function under fourier transforms, but this is a hint, and an interesting note.

 

I am sitting with my memory full of a set of distinguishable events and note a frequency distribution that is fluctuating. How can I make progress and learn? Can I somehow, use part of my memory to investigae a transformation of the fluctuations, and so to speak extract more information? I seek the transformation to use, that gives me maximum yield, given the constraints and NO PRIOR preferred symmetry of the deviations. Sometime like that... but I am still looking for formalise the question. I have a feeling that a selection will take place and the transformation which are "most fit" will ultimately come to dominate, given enough processing.

 

Somehow this is a key question, I don't find it attractive to introduce a transformation without induction. The transformation must be self-induced at some level.

 

/Fredrik

Admittedly it seems difficult for our brain to objectivise itself. I would hazard that we simply haven't got enough data yet. What we can see is definitely interesting, even surprising, but we need to 'dissect' things a whole lot more. Much of the historical record comes from the study of malfunctioning or damaged brains -we're only starting to see more detail, more levels, and mapping different functional areas. The brain appears to be a multiple kind of structure (unlike, say, the liver which is fairly homogenous).

I still think this particular NP-hard problem is waiting for the technology...