The blinking spotlight of attention

1
The blinking spotlight of attention
Rufin VanRullen, Thomas Carlson, and Patrick
Cavanagh PNAS December 4, 2007 vol. 104
no. 49
2
Attention to a single or to multiple objects
1- Parallel and always continuous 2-
Sample-when-divided epoch-wise to multiple
objects but continuous to a single object 3-
Sample-always Epoch-wise even on a single
object resetting of epochs Parallel and
epoch-wise Possible and mentioned in discussion
but not tested
3
Attention to a single or to multiple objects
1- Parallel and always continuous 2-
Sample-when-divided epoch-wise to multiple
objects but continuous to a single object 3-
Sample-always Epoch-wise even on a single
object resetting of epochs Parallel and
epoch-wise Possible and mentioned in discussion
but not tested
Experiment A task that performance depends on
stimulus duration to see the effect of epoch
resetting in signle and multiple target
situation.
4
Experiment detection of contrast decrease
Weak stimulus Difficult task
OR
Strong stimulus Easy task
Performance depends on stimulation duration.
5
Experiment detection of contrast decrease
Weak stimulus Difficult task
OR
Strong stimulus Easy task
Performance depends on attention condition.
6
Models and parameters
division cost
Epoch duration
7
Models and parameters
The divided attention curve is a weighted
average of full and minimal attention. Weight
depends on the set size and division cost.
Features The second derivative is negative in
half of the curve and it asymptotes to
probability of 1. Divided attention curve is
always between full and minimal.
division cost
8
Models and parameters
Several hypothetical data points based on
different values of parameters? a reference data
pool.
In sample-always, the full-attention is not
sigmoid
division cost
9
Empirical data collection and analysis
Compare the sigmoid Compare lt100ms data
Root Mean Square Distance of the empirical data
points from the model-generated data
Easy? parallel model Difficult? sample-always
10
Empirical data collection and analysis
Compare the sigmoid Compare lt100ms data
Root Mean Square Distance of the empirical data
points from the model-generated data
Easy? parallel model Difficult? sample-always
11
Experiment 2
Push-cue instead of Pull-cue Reaction Time
instead of Error Rate Exposure time determined by
RT of subject instead of by the task
The RT plot shows parallel strategy for the easy
task but serial strategy for the difficult task.
The ER plot shows that the RT is RT for
maintaining the same ER therefore fluctuations
in ER are not confounding.
12
Detection theory insightful hit, accidental hit,
miss
The frequency of accidental hit is the same as
that of false alarm
13
(No Transcript)
14
Scaling to (1-fs) above is similar to what we did
before
?