Nate Silver's book contains an interesting observation: human's predictive ability (pattern recognition / intelligence) is a succession of approximations.
how do humans recognise a cat? construct internal representation (2D/3D), look for distinguishing features, match against known.
What about a software library that does this?
Video of cat
-> basic 3D skeleton of cat
-> 3D animated model of cat
-> average position of model gives static model
-> shape simplification until "matches" a shape from the database? hash? similarity tree?
-> choose a part (e.g. head) then simplify until matches a hash?
Test on pictures of clouds.
Possible that some facial recognition technologies already do something like this?
