Perception and Cognition Lab UCSD John Serences
<p>Feb 10, 2010 8:03 PM
 
JOHN SERENCES UCSD Asst Prof.  I hear today at MBC ... v smart.
Bill Newsome in aud. v impressed
 
PAYING Attention to the Most Informative Sensory Neurons
 
Behavior Goals Influence Sensation, Memory, and Decisions.
He shows video of taxi cabs in Times Square...  what we pay attn to is heavily influenced
by our goals/ interests.
 
Shows ideal neuron that spikes 60x/sec in response to line segment at 90 degr orientation and
falls off as a bell curve with other orientations...  also  its noisy and therefore hard to
distinguish line segments that are at 85 degrees etc.
 
If we look at the avg of whole population of neurons, do better (? cuz of law of large numbers).
 
In 1985 Moran and Bob Desimone (DES ee moan) showed that attention biases
spiking rate.  eg  with green segment that is at 90 degr to red segment (at 135).
If subject is told to pay attn to green segment then THOSE neurons have a higher
spiking rate at 45 degr than the red neurons do the red line segment at 135.
 
HIS CONTRIBUTION:  to do fine discriminations (really close distractors)
actually (unlike Desimone) the gain should  be in OFF-CHANNEL neurons
to INCREASE the difference btwn the stimulus and the distractor.
 
He shows subjects very brief (eg 50 msec)  exposure of grating oriented eg at
-45 degr.... they need to distinguish it from gratings at 50 degrees...  fine discrimination.
It is a hard task to do.  He shows that they do best when their OFF-CHANNEL
gain is maximized  (ie NOT precisely at say 50 dgrees.)
 
This is a skill that eg radiologists need to distinguish tumors on XRAY.
 
Each voxel at medium resolution is perhaps  3mm by 3 mm by 3mm = 27 mm3.
Note includes huge number of neurons.
(David Brooks says its like wanting to hear dinner conversations and instead
seeing lights in a city.)
 
He describes medium resolution scan as 2mm by 2 by 1.5 mm voxels.
 
Can get a TUNING response in individual voxels in visual areas... easiest (most S/R)
in V1.   Y axis is BOLD signal and X axis is line segment orientation.
 
Shos subjects Tic tac toe  grid at 45 degr orientation...  and tell them to
either focus on R sloping or L sloping lines and now thicken one or the other
and have them guess which one was thickened.
Decr contrast until they can just barely notice the difference  to measure the gain.
 
?  relation of BOLD to single unit firings.
try to get improved encoding capacity by incr dynamic range of response.
 
Off-Channel  neural gain  DOES predict performance on this task.
ie the subjects with best off-channel neural gain ARE the ones that can the task the best.
 
Summary:  the TASK or GOAL of a behavior is a crucial determinant of attention/ perception.
 
 </p>