Tuesday, 08 May, 2001
autoassociators can be translation-invariant at recognising objects when they are moved about against a blank background
but when you have a second object, the autoassociator is a point attractor and eventually chooses one of them
similarly, a complex background with lots of objects: it gets confused
receptive fields aren�t static � they get smaller, the more complicated the background or the more objects in the visual field/receptive field
two types of attention: spatial and object
what can a Boltzmann machine do that a normal autoassociator can�t???