Greg Detre
@2 on Friday, 26 May, 2000
Neural computation labs, EP
self-organising
1 dimensional (1 degree of freedom)
gaussian (i.e. bell curves sensitivity of response) in a ring � each numbered neuron represents x� in one direction from the next
continue to function even when visual input ceases, i.e. in the dark (so gets input from motor, and vestibular too probably)
do the head position neurons specify an absolute position in space?
effectively yes, if combined with information about space (e.g. the place cells), right?
specify absolute location in 2-D overhead space
been found in rats
integrate the 2 to form an agent who knows about his position, even when moving in the dark � forms an internal representation of his spatial environment
Objects against a cluttered background � works in the testing phase
gave me 3 papers (original Visnet paper, cluttered environment, new learning rule)
been done before � useful as a front end
isn�t embodied, so cannot correlate motor input with 3D sensory input changes
position invariance (unless trained with the position-varying visual primitives/elements, as with the rotation invariance spherical projection visual elements)
can�t generalise to �face� in general, on the basis of the 7 facial stimuli it�s been shown
and probably can�t cope with remembering many faces, because of interference between them
perhaps because it�s only got 1000 neurons in total in each layer (4 of them)