Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
B. Liu, M.V. Macellaio and L.C. Osborne Rapid gain adaptation optimizes pursuit accuracy
Department of Neurobiology, The University of Chicago, Chicago IL, USA
2. Rescaling of response gain with stimulus variance in MT neurons and pursuit:
In the natural world, the statistics of sensory stimuli fluctuate across a wide range. In theory, the brain could maximize information recovery if sensory neurons adaptively rescale their sensitivity to match their limited response bandwidth to the current range of inputs. Such adaptive coding has been observed in a variety of systems, but the premise that adaptation optimizes behavior has not been tested. Here we show that adaptation in cortical sensory neurons maximizes information about visual motion, and minimizes tracking errors in pursuit eye movements guided by that cortical activity. Thus efficient sensory coding is not simply an ideal standard but rather a compact description of real sensory computation that manifests in improved behavioral performance.
purs
uit
MT
neur
on
a
b
∆I (b
its)
Imean (bits)
n=230 0.5 1.5
-0.9
-0.6
-0.3
0
1
c
d
time (ms)-100 0 100 200 300
info
mat
ion
(bits
)
0
0.2
0.4
0.6
0.8 datashuffled
-100 0 100 200 300time (ms)
info
mat
ion
(bits
)
0
0.1
0.2
0.3
0.4 datashuffled
∆I
normalized stimulus10.50-0.5-1
norm
aliz
ed re
spon
se
0
0.5
11/81/41/21248
e
f
0.125 1 8gain scale factor
norm
aliz
ed in
form
atio
n
MT (n=94)pursuit (n=7)0
0.20.40.60.8
1
3. Mutual information: Maximize information
by adaptive gain in MT neurons and pursuit.
Rapid recovery of mutual information after step change in direction variance.
prob
abilit
y
d
0102030
a
eye direction (deg)
targ
et d
irect
ion
(deg
)
0
0.03
0.06
0.09
0.12-30
-10-20
-3 -1.5 0 1.5 3
H
400 80 120160200
prob
abilit
y of
tria
ls
0
0.15
0.3
0.4572ms HTL
n=6504
time after variance shift (ms)
c
f
e
40 80 120160200
prob
abilit
y of
tria
ls
54ms
LTH
n=6066
0time after variance shift (ms)
0
0.15
0.3
0.45
Prob
abilit
y of
tria
ls
0
0.2
0.4
0.6
44ms
LTH
n=13623
time after variance shift (ms)40 80 1201602000
n=13744
prob
abilit
y of
tria
ls
0
0.2
0.4
0.6 61msHTL
time after variance shift (ms)40 80 120160 2000
6. Variance shift detection on single trials:MT neuron
pursuitLTH HTL
b
0 100 150 200
thre
shol
d
50
trials
time after variance shift (ms)
threshold
0
5
10
15
1. Introduction and Experiments:
The theory of efficient coding is linked to the idea that neural systems maximize information relevant to behavioral performance that can influence survival. Observations of neural responses in many organisms have demonstrated a capacity for efficient coding, but the consequences for behavior have not been explored. In our work, we demonstrate for the first time that efficient coding applies to a neural system as a whole, improving the accuracy of the movements it generates, and not solely to individual sensory neurons. We have exploited the close connection between cortical motion estimates and smooth pursuit eye movements to demonstrate parallel adaptation effects in sensory neurons and movement behavior. We find that adaptation to motion variance optimizes the encoding of motion information by MT neurons, with a behavioral impact of optimizing information in pursuit eye movements, minimizing visual tracking errors, and thereby improving vision of moving objects.
7. Discussion:
Funding: Alfred P. Sloan Foundation ; Whitehall Foundation ; Brain Research Foundation ; NEI EY023371 1. Barlow HB, 1961. 2. Osborne LC and Lisberger SG, 2009 3. Osborne LC et al., 2005. 4. Fairhall AL et al., 2001. 5. Brenner et al., 2000. 6. Wark B et al., 2009. 7. Wark B et al., 2007.
0 0.2 0.4 0.6 0.8-0.9
-0.6
-0.3
0
∆I (b
its)
n=11
Imean (bits)
velo
city
(deg
/s)HTL
firin
g ra
te (s
pike
s/s)
time (ms)
0
30
0 500 1000 1500
HTL
hvel
vvel
0
60
time (ms)0 500 1000 1500
stim
ulus
(deg
)200
00 1000 2000 3000
time (ms)
150
0
time (ms)0 1000 2000 3000
dots on
MT neuron recording eye tracking
stim
ulus
(deg
)
SD (d
eg)
SD (d
eg)
MT
neur
onpu
rsui
t
b
a
60300-30-60-100
0100200300400
L (SD 12°)H (SD 35°)
direction (deg)
rate
(spi
kes/
s)
L
H
-20-80 -40 0 40 80
eye
dir (
deg)
-10
0
10
20
L (SD 5°)H (SD 15°)
L
H
direction (deg)
rate
/mea
n
420-2-4
0
2
4
6
L(SD 12°)H(SD 35°)
direction/SD
c
direction/SD-8 -4
eye
dir/m
ean
-4
-2
0
2
4
L(SD 5°)H(SD 15°)
0 4 8
d
e
f gain indexnorm indexfilterednorm filtered
0
0.2
0.4
0.6
∆g/<g>-0.5 0 0.5 1-1
fract
ion
0
0.1
0.2
0.3
0.4
∆g/<g>-0.5 0 0.5 1-1
gain indexnorm indexfilterednorm filtered
fract
ion
time (ms) 0 100 150 200
erro
r (de
g)0
4
8
12datano adapt
min error
50direction error (deg)
-30 -15 0
prob
abilit
y
00.10.20.30.40.5
datano adapt
min error
15 30
b ca
0 0.5 1.5 21gain scale factor
rms
erro
r (de
g)
0
6
12
18 SD 1° 2.5°5°10°15°20°25°30°36°40°45°
d e f
0target SD (deg)10 20 30 40 50
∆erro
r (de
g)0
0.51
1.52
2.5
erga0
120
0
150
0target SD (deg)10 20 30 40 50
purs
uit e
rror (
deg)
no adapt modeldatamin error
er
ga
dire
ctio
n (d
eg)
-30
-15
0
15
30
stimulusdatano adapt
time (ms) 0 100 150 20050
4. Optimal gain for minimizing pursuit error:
5. Gain adaptation depends on experienced stimulus values:
-60 -30 0
L
TH
-1000
100200300400
direction (deg)30 60
firin
g ra
te (s
pike
s/s)
a c e
gain
(spi
kes/
s de
g)
direction SD (deg)10 20 30 40
0
25
50
L to TL to H
f
LTH 1 2
time (ms) -100 -50 0 50 100
dire
ctio
n (d
eg)
-60
-30
0
30
60
prob
abilit
y
direction (deg)
H-60 6020-20
T
b
L
direction (deg) -80 -40 0
eye
dire
ctio
n (d
eg)
-20
-10
0
10
20
40 80
L
TH
d0.6
n=11
L to TL to H
gain
0
0.2
0.4
normalized SDL H
MT neuron
pursuit
Barlow’s hypothesis of efficientsensory coding:
−60 −30 0 30 600
0.05
0.1
prob
abilit
y
−60 −30 0 30 600
0.5
1
stimulus value (s)
resp
onse
(r)
stimulus distribution
high
gai
n
low gain
threshold