Carnegie Mellon University, Pittsburgh, PA
description
Transcript of Carnegie Mellon University, Pittsburgh, PA
V1 interactions reduce local uncertainty about binocular disparity over time
Jason M. Samonds, Ben Poole, Tai Sing Lee
Carnegie Mellon University, Pittsburgh, PA
V1 perspective of incoming visual information
Noise and background add uncertainty
Uncertainty in corresponding features between the eyes
left eye right eye
percept
stereo correspondence problem
Computer vision solution: sharing information
Recording & Stimulation
100μV
0.2ms
5mm
Medial - Lateral
Pos
terio
r - A
nter
ior
1°
Samonds et al., J Neurosci (2009)
Recording & Stimulation
A
P
1°
Disparity Tuning Similarity (rdisp)
ΔR
F D
ista
nce
(°) r = 0.32
p = 0.01
0.0
0.5
1.0
1.5
-1.0 -0.5 0.0 0.5 1.0
A n = 63 Pairs
Evidence of disparity-dependent neuronal interactions
Samonds et al., J Neurosci (2009)
Evidence of disparity-dependent neuronal interactions
Disparity Tuning Similarity (rdisp)-0.1
0
0.1
0.2
0.3
-1.0 -0.5 0.0 0.5 1.0
Spik
e C
orre
latio
nr = 0.50p < 0.001
0.4
Samonds et al., J Neurosci (2009)
Binocular disparity tuning sharpens over time
Binocular Disparity (±1°)
Firin
g R
ate
Response Onset End of Stimulation
Samonds et al., J Neurosci (2009)
How do we quantify sharpening?
0
25
50
75
100
0 500 1000
Firin
g R
ate
(sps
)
preferred disparity2nd preferred3rd preferredleast preferred
Time (ms)
Measure sharpening with skewness
3
1
2
1
3
33
1
))((1
))((1
N
d
N
d
fdfN
fdfN
skewness << 0
skewness << 0
skewness = 0
skewness = 0
skewness >> 0
skewness >> 0disparity
meanfiring rate
firin
g ra
teSamonds et al., J Neurosci (2009)
Measure sharpening with skewness
skewness = 0.5 skewness = 2.0
Measure sharpening with skewness
0.0
0.5
1.0
1.5
2.0
2.5
Skew
ness
1.5
0.0
0.3
0.6
0.91.2
Skew
ness
200 400 600 800 1000Time (ms)
-100 00.0
0.3
0.6
0.91.2
1.5
Skew
ness
200 400 600 800 1000Time (ms)
-100 0
Measure sharpening with skewness
0
0.2
0.4
0.6
0.8
-100 0 100 200 300 400 500 600 700 800 900 1000
Time (ms)
Skew
ness
0
10
20
30
40
50Fi
ring
Rat
e (s
ps)
n = 41 neurons
-100 0 100 200 300 400 500 600 700 800 900 1000
Time (ms)
Increasing uncertainty – reduced binocular correspondence
100% correspondence 50% correspondence
Increasing uncertainty – reduced binocular correspondence
-0.20.00.20.40.60.81.01.21.4
0 200 400 600 800 1000Time (ms)
Skew
ness
n = 14 neurons
0
10
20
30
40
50
0 200 400 600 800 1000Time (ms)
Firin
g R
ate
(sps
) 100% correspondence 50% correspondence
0
10
20
30
40
50
0 200 400 600 800 1000Time (ms)
Firin
g R
ate
(sps
)
-0.20.00.20.40.60.81.01.21.4
0 200 400 600 800 1000Time (ms)
Skew
ness
A B A B A B B A B A B A
Ambiguous Stereogram
0°A B A B A
B A B A B
-0.188°
+0.188°
left eye right eye25% unambiguous random dots
Increasing uncertainty – correspondence ambiguity
Julesz & Chang, Biol Cybern (1976)
1°
0
10
20
30
-1.0 -0.5 0.0 0.5 1.0
Horizontal Disparity (°)
Firin
g R
ate
(sps
)
25% unambiguous random dots
1°
Increasing uncertainty – correspondence ambiguity
-100 0 100 200 300 400 500 600 700 800 900 1000Time (ms)
0
10
20
30
40
50
Firin
g R
ate
(sps
)
far vs. near disparity (center)
0
10
20
30
40
50
-100 0 100 200 300 400 500 600 700 800 900 1000Time (ms)
Firin
g R
ate
(sps
)
far vs. near disparity (surround)1°
25% unambiguous random dots
1°
Increasing uncertainty – correspondence ambiguity
10
Time (ms)0 200 400 600 800 1000
0
10
20
30
40
50 far vs. near disparity (center)
Firin
g R
ate
(sps
)
Increasing uncertainty – correspondence ambiguity
0
10
20
30
40
50
0 200 400 600 800 1000
Firin
g R
ate
(sps
)
Time (ms)
n = 36 neurons
far vs. near disparity (surround)
Summary
Spike correlation in V1 depends on spatial and binocular disparity tuning relationships.
Binocular disparity tuning sharpens over time.
Binocular disparity tuning sharpening is clearer when stimuli arepresented with reduced or ambiguous stereo correspondence.
Conclusion
Recurrent interactions within and/or with V1 share binoculardisparity information across space to reduce local uncertainty within individual V1 receptive fields.
Thank you for your attention!
Brian Potetz, Karen McCracken, Matt Smith, Ryan Poplin, and Nicholas Hatsopoulos for assistance.
Supported by: NEI F32 EY017770, NSF CISE IIS 0713206, AFOSR FA9550-09-1-0678 and a grant from Pennsylvania Department of Health through the Commonwealth Universal Research Enhancement Program.
Acknowledgements