[ModelBreaker]

You should check out the work of Dr Laurent (sp?) Itty (the last name is right) of USC. I haven't updated myself on it for a couple of years. But a couple of years ago, his software pick out 'interesting' features from a moving video stream. He uses an SOM modeled after features of the human eye and it works really well.

The problem is that, as you mention, it's very very processor intensive. Many computers running in parallel, if I recall correctly.

[Nataku X]

Is it completely learning-based? As far as I know SOMs that are entirely based on learning are very limited, let alone a comphrensive, totally-learning-based inconsistency detector that incorporates orientation, motion, color, etc... a learning based approach is the only way to fully emulate the human brain... although not everything is learned (color is hard-wired within the photoreceptors, for example)...