TEMPLATE DESIGN © 2007
www.PosterPresentations.com
Scaling of anticipatory smooth pursuit eye movements with target
Scaling of anticipatory smooth pursuit eye movements with target
speed probability
speed probability
David Souto, Anna Montagnini & Guillaume S. Masson
david.souto@pse.unige.ch or anna.montagnini@incm.cnrs-mrs.fr
Background Disagreement with previous
models
References
Direction
randomization
Anticipatory smooth pursuit permits tracking of
expected target trajectories with little delay. It
can be a mirror of cognitive expectations, but It
is still unclear how these expectations are built.
Pursuit anticipation scales with several target
parameters in an adaptive way [1]. This is
compatible, for instance, with a switch between
expectation states after one or several wrong
guesses [2]. Alternatively, the underlying
mechanism could rely on a global subjective
probability estimation.
Goals – clarify the nature of the expectations
that drive anticipatory pursuit
z by examining effects of velocity and direction
randomization across blocks of different
probability
z by looking at trial history effects
Markov two-states model [2]:
Anticipatory pursuit may reflect alternation
between two mutually exclusive states of
expectation (high vs. low velocity or left vs.
rightward). This model predicts a bimodal
distribution of anticipatory pursuit for
intermediate values of p (black curve)
3 human subjects participated in two
experiments. Eye movements were recorded
with a scleral search coil.
Each probability block comprised 250-500 trials
with two possible values of target speed (Exp. 1)
or direction (Exp. 2). The probability p of the
highest speed (or of the right direction) to be
presented was varied from 0 to 100% across
blocks.
VSS 2008
@ Naples, Florida
Data analysis
Direction rand.
Trial-history effects
Speed
randomization
Conclusion
Speed rand.
5 °/s
15 °/s
V80
300-450
300ms
gap
fixation
(LED)
500 ms
P (V=15)
={0 -100%}
P (Right)
= {0 -100%}
15 °/s
t [ms]
Probability bias
http://www.unige.ch/fapse/cognition/souto
[1] Heinen, S. J., Badler, J. B., & Ting, W.
(2005). J Vis, 5(6), 493-503.
[2] Kowler, E., Martins, A. J., & Pavel, M. (1984).
Vis Res, 24(3), 197-210.
Effects of the n
th
previous trial on the
current trial, for a 5
(upper panels) or 15
deg/sec target
motion.
Error-bars = C.I.
Effects increase with
scarcity of stimulus
Sequence effects
Speed rand. Direction rand.
Distribution
Speed rand.
Direction rand.
z Scaling of anticipatory speed with probability:
favoring low speeds for speed randomization,
nearly linear for direction randomization
z Large effects of the last 2-4 trials. Two wrong
expectations are not enough to turn off
anticipations in biased blocks.
z Anticipation may better be explained by a
continuous accumulation-of-evidence model
z Coming soon: quantitative predictions
Scaling with block probability
Anticipatory velocity
(V80), latency and
mean anticipatory
acceleration scale
with probability bias
Anticipatory velocity
(V80) scales linearly
with probability bias.
Anticipatory pursuit
onset and mean
acceleration scale
with absolute
anticipatory velocity.
anticipatory
pursuit onset
p(Left) = .75
Effect of n
th
previous trial on
current trial when
it was a leftward
or rightward
target.
Error-bars = C.I.
Anticipatory velocity (V80) has a
unimodal distribution which shifts with p
both for speed and direction
randomization
velocity (deg/sec)
2 4 6 8 10
-1
-0.5
0
0.5
1
trials back
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
p = .10 p = .50
-1
-0.5
0
0.5
1
p = .25 p = .75 p = .90
p(V=15)
n-trial = 5 deg/sec
p = .10 p = .50 p = .25 p = .75 p = .90
n-trial = 15 deg/sec
Observer AM
-0.5
0
0.5
1
1.5
2 4 6 8 10
-1.5
-1
-0.5
0
0.5
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
p(Right)
n-trial = right
Observer AM
n-trial = left
trials back
velocity (deg/sec)
p = .10 p = .50 p = .25 p = .75 p = .90
p = .10 p = .50 p = .25 p = .75 p = .90
36.535
500 ms
Our experimental results (red curves) disagree
with this prediction: an alternative model could
better predict the shift of the unimodal
distribution with p (e.g. a continuous
accumulation-of-evidence model)
p(Right) = .25
-15 -10 -5 0 5 10 15
0
0.2
0.4
0.6
0.8
1
anticipatory velocity at V80 (deg/s)
P(X<x)
0
10
25
50
75
90
100
P(Right)
AM
-8
-6
-4
-2
0
2
4
6
8
anticipatory
velocity at V80
(deg/s)
-250
-200
-150
-100
-50
0
anticipatory pursuit
onset (ms)
0 25 50 75 100
0
5
10
15
20
25
30
anticipatory
acceleration at V80
(deg/s)
0 25 50 75 100 0 25 50 75 100
Probability of Right-direction (%)
AM DS AR
AM DS AR
AM DS AR
25 50 75
-8
-4
0
4
8
25 50 75 25 50 75
25 50 75
0
2
4
6
8
10
P(V=15)
25 50 75 25 50 75
v5-v5 sequence
v15-v15 sequence
LL sequence
RR sequence
AM DS GM
AM DS AR
P(Right)
ant. vel (V80) [deg/s]
ant. vel (V80) [deg/s]
-10 -5 0 5 10
0
0.2
0.4
0.6
p =0
-10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10
-10 -5 0 5 10
0
0.2
0.4
0.6
-10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10
-10 -5 0 5 10
0
0.2
0.4
0.6
-10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10 -10 -5 0 5 10
ant. vel. at v80 (deg/s)
FSMM model prediction
experimental data
DS
AM
AR
p = 10 p = 25
p(Right)
p= 75 p= 90 p = 100
ant. vel. at v80 (deg/s)
ant. vel. at v80 (deg/s)
p = 50
density
Pursuit task
-5 0 5 10 15
0
0.2
0.4
0.6
0.8
1
0
10
25
50
75
90
100
P(V=15)
anticipatory velocity at V80 (deg/s)
P(X<x)
AM
0
2
4
6
8
10
-250
-200
-150
-100
-50
0
20
40
60
80
100
120
AM DS GM
AM DS GM
AM DS GM
anticipatory
velocity at V80
(deg/s)
anticipatory pursuit
onset (ms)
anticipatory
acceleration at V80
(deg/s)
Probability of High Velocity (%)
0 25 50 75 100 0 25 50 75 100 0 25 50 75 100