Neat quote: ``I've reluctantly discarded the notion of continuing to manage the portfolio after my death - abandoning my hope to give new meaning to the term `thinking outside the box.''' --- Warren Buffett.

Anyways, some research rant (don't expect this to make sense---it doesn't fully make sense to me yet): It may be possible to use a neural network to process and store images that the `brain' sees. A way to overcome the scale/rotation/perspective problems is to use an extra layer of a neural network that represents the scale/rotation/perspective matrix. I've considered this before, but then you'd still be struck trying to figure out -which- scale/rotation/perspective matrix to use (ie: what values to put in there). For example, you see a car 100 feet away, and recognize it as a car. then you see the same car 10 feet away from a different angle, and you still recognize it. A single matrix won't let that happen. So the trick is to notice the parallel nature of the brain. Don't think of a matrix, think of all the possible matrices! Heck, have a billion of them just for this task alone. Each one encoding a slightly different variation on the matrix values. But we're still left with the problem of having to implement this on a computer, and the computer will still need to do all the computations, but what if some representations become more common than others? What if the more a certain matrix is used, the more likely it is to be used again? For example, most human faces we see are `properly' oriented---so your brain doesn't need to bother looking through the space of all possible matrices, just the ones that are oriented right. There may be a way of sampling this space. More on this later; need to let this idea cook.