Statistical Models and Sensory Attention
Peter Dayan Rich Zemel
In ICANN, 1999, 1017-1022.
Abstract
Physiological investigations into the neural basis of sensory
attention have led to puzzling and contradictory results. Attention
can seemingly lead to increased, decreased and unchanged neural
activities, according to features of attentional experiments that
are not well understood. We take one particular case in which
activities increase as a result of attention, model its possible
statistical underpinning, and relate our model to other
attentional suggestions. Increased activities in population codes
are associated with increased certainty about the encoded
quantities. This increased certainty has to come from somewhere --
in our model it emerges from particular changes in the model's
processing strategy.
compressed postscript
pdf