[MagnusWootton]

PPL all think you cant optimize nearest neighbour past looking at every element in the memory, for every access element, before you can have a solution.
But I have a possible solution (I havent coded it yet tho...) but it has to be photo keys (graphical keys) and it only finds the nearest neighbour not below it.

Its a big winner,  and ppl will be helped alot by knowing this.

So it goes like this->
    * form a mipped power of 2 heirarchy of averagations of the final photo patches,  and do this also to the photos your using to access the content addressable memory.

  * then you access blurry first,  and each level, you take off all the keys that werent the highest scoring, for each access.   Then you dont need to access all the high resolution keys, if the blurry ones didnt also match the highest on their tiers.

So that way, it is after all simply spacially dividable to do it,  and its not insane genius.     Also It'll get an amazing video tracker to work, maybe even on an arduino!
It could work on general data keys,  but it has to be true to a blur, if it works or not, then its ok to try on anything.

Because neural networks are very similar to KNN,  it also optimizes them as well,  this is quite a dangerous thing if its true.    heheh.