4:15pm
Location:
Munzer Auditorium,
Title: "Revisiting Population Codes"
Alexandre Pouget
University of Rochester
Abstact:
Many studies have attempted to relate changes in
behavioral performance (e.g. as a result of perceptual learning or increased
attention) to changes in neural codes. One general principle has emerged from
these studies: the steeper (or the narrower) the tuning curve, the better the
performance, or equivalently, the higher the information. This principle,
however, is based on the assumption that increasing the slope of the tuning
curves does not affect the noise distribution. We will show that in networks of
spiking neurons, this assumption does not necessarily hold. In many networks,
an increase in the slope of the tuning curve leads to changes in the noise
distribution, such as the overall information decreases. We will illustrate
this problem in the context of orientation selectivity. We will show that
sharpening tuning curves through lateral connections result in very large
information losses and lead to complex codes. We will discuss the implications
of those findings for the neural basis of perceptual learning and attention.