learning and tuning of neurons in inferior temporal cortex
DESCRIPTION
Learning and Tuning of Neurons in Inferior Temporal Cortex. Learning and Neural Plasticity in the Adult Visual System Society for Neuroscience San Diego, California. Bharathi Jagadeesh Department of Physiology & Biophysics University of Washington Seattle, Washington. Macaque temporal lobe. PG. - PowerPoint PPT PresentationTRANSCRIPT
Learning and Tuning of Neurons in Inferior Temporal Cortex
Bharathi JagadeeshDepartment of Physiology & BiophysicsUniversity of WashingtonSeattle, Washington
Learning and Neural Plasticity in the Adult Visual SystemSociety for Neuroscience
San Diego, California
V1
PGPF
V3
V1 V2 V4 TEO TE
TE
Ventral or “What” processing stream
Macaque temporal lobe
Pictures of people, places, and things
100
s/s
300 ms
best next best
worst
What does the selectivity in IT mean?
V1 V2 V4 TEO ITre
spo
nse
Perceptual similarity
IT
Perceptual similarity
• Image characteristics
• Experience
Perceptual similarity
Neural responses in IT
Proposed relationship
Similarity of stimuli should explain selectivity in IT cortex
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
Proposed relationship
Measuring perceptual similarity
How do we use perceptual similarity algorithms?
Image similarity algorithms
Wang et al (2001)
SIMPlicity algorithm
1. Image divided into 4x4 pixel blocks, feature vector is calculated for each block.
2. Feature vector 6-dimensional: Color dimensions, (LUV space, 3 dimensions) , Spatial frequency, wavelet analysis on the L component of the image (3 dimensions)
3. Number of regions using a k-means algorithm.
4. The similarity between two images computed by comparing regions using Integrated Region Matching (IRM).
5. The SIMPLIcity (similarity) distance is weighted sum of similarity between regions.
25.8 32.2
38.4
33.559.9
47.5
31.2
57.9
30.0
40.954.9
32.1
Calculate image distances between images
Prediction
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
Proposed relationship
?
3.54 4.07 4.07 4.34 4.92
5.13 5.24 5.56 5.72 5.90 5.93
6.05 6.39 6.42 6.49 7.00 7.12
7.33 7.34 7.49 7.50 7.50 7.51
SIMPlicity retrieves targets
50 100 150 200
0.25
0.5
0.75
1.00SIMPLIcity
Number of Relevant Images Retrieved
Pre
cisi
on
Algorithms predict perceptual similarity
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
Proposed relationship
?
Do perceptual similarity algorithms explain neural responses in IT cortex?
100
s/s
300 ms
best next best
worst
Example cell: image distance between best/next and best/worst
best-next best
best-worst
0
15
30
45
SIM
PLI
city
best
next best
best
worst
SIMPLIcity
(best and next best stimuli)
best
worst
next best
best
Population: Distance between Best-worst v. Best-Next Best
0 50 1000
50
100S
IMP
LIci
ty
( be
st a
nd w
orst
stim
uli)
30
40
50
best-next best-worst
Do other similarity algorithms explain neural responses in IT cortex?
50 100 150 200
0.25
0.50
0.75
1.00
Number of Relevant Images Retrieved
Pre
cisi
on
Contrast: Doesn’t retrieve targets
0.0 0.2 0.4 0.6
0.0
0.2
0.4
0.6
RMS contrast difference (best and next best stimuli)
RM
S c
ontr
ast
diff
eren
ce
(be
st a
nd w
orst
stim
uli)
And, doesn’t explain IT responses
best
worst
next best
best
0
0.1
0.2
best-next best-worst
EMD, another similarity metric: Retrieves targets
50 100 150 200
0.25
0.50
0.75
1.00
Number of Relevant Images Retrieved
Pre
cisi
on
And, also explains IT responses
0 50 1000
20
40
60
80
100
EMD (best and next best stimuli)
EM
D
(be
st a
nd w
orst
stim
uli)
best next best
20
30
40
best-next best-worst
Prediction
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
Proposed relationship
Behavior
?
Fixation250-500ms
Response
Stimulus16-512ms
Mask256 ms Delay
100-500msDelay500-1200ms
Delayed match to sample (DMS) (easy pair)
Fixation250-500ms
Response
Stimulus16-512ms
Mask256 ms Delay
100-500msDelay500-1200ms
DMS (difficult pair)
Measure “perceptual similarity”
Performance @ 50 ms stimulus presentation
Low performance High performance
71% 96%
Measure neural selectivity
Average neural response difference in passive fixation task
Performance @ 50 ms stimulus presentation
Low performance High performance
71% 96%
62% 86%
0.6 0.7 0.8 0.9 10.6
0.7
0.8
0.9
1
Behavioral performance
Neu
ral R
OC
r = 0.57
Neural performance v Behavior
Prediction
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
Individual correlations
0
0.2
0.4
Behavior v
neuron
Algorithmv
neuron
0.6C
orr
ela
tion
r
Prediction
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
Partial correlations
0
0.2
0.4
0.6
Behavior v
neuron
Algorithmv
neuron
Pa
rtia
l cor
rela
tion
r
Perceptual similarity correlated with IT neuron response similarity
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
How does training change the relationship to the algorithm?
No Go
Go
FixationPredictor
DelayChoice
Barrelease
Erickson CA, Desimone R (1999)
Passive Association Task
Go Trials No Go TrialsPredictor Choice Predictor Choice
Go 359 ms
Go377 ms
Go376 ms
Go369 ms
No Go
No Go
No Go
No Go
Valid Association Trials
Invalid Association Trials
Go 359/473 ms
Go377/465 ms
Go376/436 ms
Go369/363 ms
No Go
No Go
No Go
No Go
Go Trials No Go TrialsPredictor Target Predictor Target
Response to choice stimulus is correlated with response to predictor
Neural response to predictor
Ne
ura
l re
spo
nse
to c
hoi
ce
Erickson & Desimone (1999)
Dissimilar stimuli produce similar responses
Erickson & Desimone (1999)
Training breaks relationship to algorithm
35
37
39
41
43
Similar Responses
Differentresponses
Imag
e di
stan
ce35
37
39
41
43
Similar responses
Different responses
Imag
e di
stan
ce
Association task: TrainingPassive fixation: No training
Data from Erickson & Desimone (1999)
Specific conclusions
•Perceptual similarity algorithms measure (at least partially) the perceptual similarity of stimuli.
•The same algorithms explain (at least partially) the neural response similarity in IT.
•But, neural response similarity is better correlated with discrimination performance (measured perceptual similarity) than it is with the image similarity algorithms.
•And, training that modifies the processing of stimuli breaks the relationship between image similarity and neural response similarity.
Katie AhlSarah AllredYan Liu, M.D., Ph.D.Jen Skiver Thompson
Andrew Derrington, Ph.D.Cynthia Erickson, Ph.D.J.M.Jagadeesh, Ph.D.Amanda Parker, Ph.D.
Jamie BullisRebecca MeaseAmber McAlister
Current members
Rotation students & former members
Visiting scholars and collaborators
Jagadeesh labUniversity of Washington
Divider
Methods
Record from single neurons in the non-human primate brain, while the primate performs visual tasks.
Monitor eye movements so that visual stimulus at the retina is known.
In my lab, we record from inferotemporal cortex in the macaque, and are of cortex thought to be important for perception of people places and things.
100
s/s
300 ms
best next best
worst
B
0 50 1000
50
100
SIMPLIcity
(within effective group)
SIM
PLI
city
(eff
ectiv
e v
inef
fect
ive
grou
ps)
35
37
39
41
43
Eff Eff-Ineff
Ima
ge
dis
tan
ce
Population within v across groups
0 100 200 300 400 500 6000
10
20
30
40
time in ms after stimulus onset
spik
es
pe
r se
con
d
rank 1-8 rank 10-17rank 18-24
Histograms, population, sorted by emd rank to best
-1 -0.5 0 0.5 1
0
20
40
60
80
r-value
freq
uenc
y
Population: all comparisons
20 40 60 800
20
40
60
80
SIMPLIcity to best
ne
ura
l re
spo
nse
(sp
ike
s/se
con
d)
20 40 60 800
20
40
60
80
SIMPLIcity to worst
ne
ura
l re
spo
nse
(sp
ike
s/se
con
d)
Example : SIMPlicity correlated with neural response
best worst
0
20
40
60
80
20
40
60
80
-1 -0.5 0 0.5 10
20
40
60
80
-100 -50 0 50 1000
20
40
60
80
R-value
fre
qu
en
cyfr
eq
ue
ncy
Slope
best
worst
Population: SIMPlicity is correlated with neural response
0 5 10 15 20 25
0
0.03
0.06
0.09
0.12
0.15
Stimulus Rank (best to worst)
Ave
rag
e r
-va
lue
(a
bso
lute
va
lue
)
SIMPLIcity
shuffle
Population: correlation to each
EMD: Does retrieve images, also explains neural response
50 100 150 200
0.25
0.5
0.75
1
Number of Relevant Images Retrieved
Pre
cisi
on
avg EMD example EMDchance
0 50 1000
20
40
60
80
100
EMD (best and next best stimuli)
EM
D
(b
est
an
d w
ors
t st
imu
li)
A B
25.8 17.7
32.2 19.3
38.4 29.2
33.5 24.9
59.9 37.8
47.5 30.4
31.225.1
57.9 35.7
30.0 22.3
40.9 31.5
54.9 39.0
32.1 25.6
Individual correlations
0.0
0.2
0.4
0.6
neuron/behavior
Image sim
/neuron
Image sim
/behavior
core
lati
on
, r
Individual correlations
0
0.2
0.4
beh/neu emd/beh emd/neu
0.6
Partial correlations
0
0.2
0.4
0.6
beh/neu img/beh img/neu
Invalid v valid latencies
300
400
500
600
700
300 400 500 600 700
valid latencies (ms)
inva
lid la
tenc
ies
(ms)
Valid v Invalid trial latencies
300
400
500
600
700
300 400 500 600 700
valid latencies (ms)
invalid
late
ncie
s (
ms)
90% of trials
10% of trials
With training, IT relationship to algorithmic image similarity breaks
10 20 30 40 50 6010
20
30
40
50
60
within effective
across
0 50 1000
50
100
within effective
across
Data from Erickson and Desimone (1999)
V1 V2 V4 TEO IT
orientation
resp
on
se
V1
Proposal: IT neurons represent the perceptual similarity of stimuli.
Prediction
Perceptual similarity algorithms
Perceptual similarity
Neural responses in IT
Proposed relationship
Behavior
?
Retrieval
0
0.1
0.2
0 0.5 1
Imag
e di
ff
Characteristics of algorithms
1) Rely heavily on color content of images
2) Pairwise comparison, not a parametric space
3) Ignores “cognitive” information
But, works for realistic images, because images that are similar to one another in the real world tend to share patterns of colors.
Algorithms predict perceptual similarity
Test of algorithm in “spiked” databases.
Correlation between algorithm and human sorting of images.
Correlation between algorithm and monkey performance in discrimination task.