Transcript No Slide Title

L. Itti: CS564 - Brain Theory and Artificial Intelligence
University of Southern California
Lecture 16. Saccades 2
Reading Assignments:
The NSL Book
The Modular Design of the Oculomotor System in Monkeys
Peter Dominey, Michael Arbib, and Amanda Alexander
Supplementary Reading:
Crowley-Arbib Saccade Model
M. Crowley, E. Oztop, and S. Marmol
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
1
de la y
FEF
PP
FOn
PPc tr
ms
Peter Dominey & Michael Arbib:
Cerebral Cortex, 2:153-175
switch
Filling in the Schemas: Neural
Network Models Based on
Monkey Neurophysiology
qv
sm
vm
FEF
vs
PP
MD
VisCx
sm
C AUDATE
Vis C x
SC
CD
TH
LG N
vm
S Nr
SNR
vs
SG
sm
de la y
Develop hypotheses on Neural
Networks that yield an equivalent
functionality:
mapping schemas (functions) to the
cooperative cooperation of sets of
brain regions (structures)
FEFvs
FEFms
SC
vs
ms
qv
FOn
wta
ey e movement
FEFvs
FEFms
B rainstem
Saccade
G enerator
Retina
VisInput
Last time, we saw that…
- Double-saccade
experiments suggest direction/amplitude coding rather
than absolute target location
- Lesion/stimulation
studies suggest that the overall system still works
when either SC or FEF is missing (but not both!)
- FEF
stimulation just after presentation of a visual target (SC lesioned)
elicits a saccade towards the “fake” FEF target first
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
3
Experimental Findings
FEF  PP
- FEF, PPC  SC
- SC  saccade generator (SG)
- FEF  BG (CD and SNr)  SC
- connection
(role in disinhibition of SC for
saccades)
- Simple
saccade: study topographic relations between sensory and
motor areas
- Memory saccade: study cortical and subcortical activity that sustains
spatial memory
- Double saccade: study dynamic remapping of target location with
intervening eye movements
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
4
Basic Model Element: Layer
2D array of neurons
 de la y
FEF
PPc tr
ms
topographic correspondence from layer to
layer
switch
PP
FOn
qv
sm
vm
vs
VisCx
sm
CD
external world: 27x27
array
TH
LG N
vm
SNR
vs
sm
model retina: 9x9
layer; so, each model
neuron represents a
small population of
biological neurons
 de la y
FEFvs
FEFms
SC
vs
ms
qv
FOn
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
wta
ey e movement
FEFvs
FEFms
B rainstem
Saccade
G enerator
Retina
VisI nput
5
Visual Input
At every iteration,
eye position determines position
of 9x9 retinal window within
27x27 outside world
if eye velocity over 200deg/sec,
retinal input is reduced
(saccadic suppression)
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
6
Direct connection retinaSC
To superficial layer of SC (vs)
responsible for reflex saccades = short-latency saccades
to target which has not been recently fixated
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
7
visual pre-processing
LGN, V1, V2, V4 and MT areas
abstracted by a single layer
possible only because we have a
very coarse (9x9) retinal input
with no image noise!
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
8
quasi-visual cells in PP
Andersen et al. (1988) found
in PP cells that code for future
eye movements.
Quasi-visual because in
double-saccade task
found cells which fire at location
of second target respective
to first target, while there never
was a retinal stimulus there!
right movement field but wrong receptive field
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
9
Double Saccade Experiment
+
time
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
10
remapping
Hypothesis: occurs primarily in PP
(in reality, may occur in many regions at once,
with connections between regions serving for fine-tuning).
problem: eye velocity signals have not been found in PP.
but eye position signals have 
Dominey and Arbib’s computational hypothesis: remapping is done
such as to compensate for difference between current eye position, and
a damped/delayed eye position signal
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
11
frontal eye fields
Bruce & Goldberg (1984): FEF contains:
- visual cells (vm)
(receive input from PP)
- movement cells (ms)
(fire just before saccade)
- visuomovement cells (sm)
(memory: fire during delay
in memory saccade task)
- postsaccadic cells
PPctr: active as long
as fixation cross present
(inhibits eye movements)
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
FOn = fixation is on
12
superior colliculus
-Input from retina (reflex saccades)
-Input from PPqv  SC qv layer
(yield saccades when FEF
lesioned)
-Inputs from FEF
How can we choose?
WTA array: saccade to
currently strongest target
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
13
basal ganglia
SNr provides tonic inhibition
of SC and thalamus, unless
prevented to do so by FEF
(directly or via CD)
Goals:
- prevent saccades while
a target is being fixated
- memorise location of
future target in memory
saccade task
FEF can selectively control the targets for saccades, overriding collicular
attempts to initiate saccades to distracting peripheral targets
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
14
The Full Dominey Model
DCEP
trig
SC
PN
RI
dimension
Retina,
VisCx
Delay
(d1)
Mechanism A
FEF
velocity
EBN
LLBN,
MLBN
TN
MN
7a/LIP/PP
Brain Stem – Saccade Burst Generator
DCEP-Damped Change in Eye Position
7a/LIP-Oculomotor Region of Posterior Parietal Cortex
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
15
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
16
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
17
Laurent Itti: CS564 - Brain Theory and Artificial Intelligence. Saccades 2
18