* [email protected]
A matched filter mechanism of visual short-term memory in macaque parietal area 7a
Chris Klink
[1,2]
1,2
2
4
3. Performance
1,2,5,6
Option 2: Matched filter mechanism
The sample stimulus evokes a rapid change in synaptic weights across the
local network which reflects its features. The response to the test-stimulus
is multiplied by these synaptic weights, yielding a measure of similarity
between sample and test stimulus (Sugase-Miyamoto et al., 2008).
Here, we test these two hypotheses against single unit activity in macaque
parietal area 7a, recorded during a spatial delayed match-to-sample task.
1 dot
2 dots
0.50
0
(MATCH)
Correct
1 dot
2 dots
500
420
0
20 30 40 50 60
(MATCH)
ΔPositionS-T (deg)
T
Match: Error
Position tuning curves
Sample
untill
response
ΔPositionS-T (deg)
Correlation in tuning between phases
Test
Delay
Cell #2
0
D
60
1 dot position 2 dots positions
60 Example Cell #2
20
*p<.05
0
200
500
-200
0
1 dot position 2 dots positions
200 400
Time (ms)
• Brief early Match-enhancement followed by Match-suppression and late second Match-enhancement
• Activity on false positives is like Match, not Non-Match
• Activity correlates with behavioural response
time
8 stimulus positions
(dots)
1 or 2 dots
No corr.
0.50
H
C
T
A
M
NON-
MATC
H
Same dot position(s)
e
pl
m
a
S
Respond (hand movement)
Reward if correct
0.65 n=78
1
D
t
s
Te ms)
t
s
Te
y
a
l
e
No corr.
0.50
2
00
(0
t
s
Te ms)
4
0
0
(2
00
m
a
S
e
l
p
D
y
a
l
e
st )
e
T ms
t
s
e
T
00
2
(0
st )
e
T ms
4
0
0
(2
00
Delay vs. Test
Sample vs. Test
(Wilcox. signed rank, p<.01)
*49/78
Med: 0.26
0
1
(Wilcox. signed rank, p<.0001)
RS-T Match trials
1
1
*59/78
Med: 0.27
0
(Wilcox. signed rank, p<.01)
*68/78
Med: 0.31
0
*68/78
Med: 0.31
*57/78
Med: 0.30
0
Delay-Test vs. Fix-Sample
0
RD-T Match trials
1
*47/78
Med: 0.23
0
RF-S Match trials
1
Other dot position(s)
Monitor
Task
• Fixate for 400 ms to start stimulus presentation
• Sample stimulus presented for 500 ms
- 1 or 2 dots, each placed at 1 of 8 possible positions
• Remember sample stimulus during a delay period
- Unpredictable delay period duration between 700-1,000 ms
• Compare a test stimulus against the sample stimulus kept in memory
• Indicate Match or Non-Match with left/right hand movement
- Juice reward if response is correct
•
•
•
ROC analysis for different trial phases on Position and (Non-)Match discriminability
- Strong Position discriminability for 1 dot (less during delay), weaker for 2 dots
- Strong (Non-)Match discriminability for 1 & 2 dots
Correlate Position and (Non-)Match discriminability within neurons
- No correlation test-test or delay-test, for either test-phase (both 1 & 2 dots) [all p> .2]
Selective delay activity is present, but does not seem to help solve the VSTM task
8. Variability
Delay
Test
Mean-matched
Fano factor (σ/μ)
2.6
Recording
• One rhesus macaque
• 117 neurons
• Parietal cortex, area 7a
- Visual response (Mountcastle et al., 1981)
- Delay activity (Constantinidis & Steinmetz, 1996)
2 dots: Position
RD-T Match trials
700 - 1000 ms
1
• Most neurons are significantly tuned for dot position(s) in
both the sample & test phase (67/78, 85%)
• Some neurons (e.g. cell #1) also demonstrate consistently
tuned sustained delay activity (40/78, 51%)
• Others (e.g. cell #2) are only tuned during sample & test
RD-T Non-Match trials
400 ms
Delay (D)
until response
Match vs.
2 dots: Non-Match
RS-T Non-Match trials
Fix (F)
0.65 n=78
RSample-Delay
0
7. Correlated noise
AUC(med)
500 ms
Test (T)
AUC(med)
Sample (S)
1 dot: Position
RS-Tn.s.
RS-D*
0
0
RS-T*
RS-Dn.s.
6. ROC analysis
Match vs.
1 dot: Non-Match
Cell #1
1
100 Example Cell #1
20 30 40 50 60
• Performance is comparable for 1 and 2 dots
• Performance is worse for smaller spatial differences
- More errors & slower correct responses
• False positive responses are approximately as fast
as correct Match-responses
- Response speed reflects monkey’s choice
2. Methods
S
100
Error
1 dot
2 dots
Non-Match: Correct
Spikes/s
0.75
Match: Correct
5. Selectivity
Spikes/s
Option 1: Tuned delay activity
Selective sustained neural activity between sample and test moments
maintains a trace of (features of ) the stimulus in memory (Funahashi et al., 1989).
580
Response time (ms)
Proportion Correct
1.00
[6]
4. Neuronal responses
Spikes/s
How is visual short-term memory (VSTM) encoded in neural activity?
(task: compare a test-stimulus to a previously seen sample-stimulus kept in memory)
Fixation
2
Div. Pharmacology, Utrecht University | 2Helmholtz Inst., Utrecht University | 3Neuromodulation & Behaviour, Netherlands Inst. for Neuroscience | 4Wageningen Res. Ctr., Wageningen University | 5Donders Inst., Radboud University | 6Biomedical Signals & Systems, University of Twente
1. Introduction
Area 7a
[5]
RSample-Test
1
[4]
, Anna Oleksiak , Albert Postma , Ineke van der Ham , Martin Lankheet , & Richard van Wezel
1,2,3
Receptive
Field
[3]
(untill response)
1.9
Match
1.2
Non-Match
-350
0
500
Time (ms)
• Time-resolved mean-matched
Fano factor (Churchland et al. 2010)
- Variability measure (σ/μ)
• FF decreases during delay
• Steep FF decline when test starts
• Test phase variability lower for
Match than for Non-Match trials
- Synaptic adjustment?
• Correlate single trial activity fluctuations (noise) around the mean within neuron
- Fixation (200 ms before Sample), sample (0-200 ms), delay (last 200 ms), test (0-200 ms)
• Noise correlations between sample & test; stronger for Match than for Non-Match
• Noise correlations between delay & test; stronger for Match than for Non-Match
• In Match trials, noise correlations between delay & test are stronger than baseline (FixSample); this effect was absent for Non-Match trials (p>.04)
9. Conclusions
Delay activity in 55% of area 7a neurons is selectively tuned for
stimulus position on a VSTM task. This selectivity is however not
correlated with the neuron’s capability of discriminating between Matches & Non-Matches
Correlated noise fluctuations in the delay and test phases of the
task and the dynamics of response variability together suggest
a matched filter mechanism of VSTM in area 7a neurons
10. References
Churchland et al. (2010). Stimulus onset quenches neural variability: a widespread
cortical phenomenon. Nature Neurosci, 13(3), 369–378
Constantinidis & Steinmetz (1996) Neuronal activity in posterior parietal area 7a during the delay periods of a spatial memory task. J Neurophysiol, 76(2), 1352–1355
Funahashi, Bruce, & Goldman-Rakic (1989). Mnemonic coding of visual space in the
monkey’s dorsolateral prefrontal cortex. J Neurophysiol, 61(2), 331–349.
Mountcastle, Andersen, & Motter (1981). The influence of attentive fixation upon the
excitability of the light-sensitive neurons of the posterior parietal cortex. J Neurosci,
1(11), 1218–1225
Sugase-Miyamoto et al. (2008) Short-term memory trace in rapidly adapting synapses of inferior temporal cortex. PLoS Comp Biol, 4(5), e1000073
CK is supported by the NCU Focus and Mass program of Utrecht University. This study
was further supported by a grant from the Netherlands Organisation for Scientific Research (NWO) and a High Potential grant from Utrecht University