a reinforcement learning model of song acquisition in the bird9.54/fall14/slides/class19_fee.pdfthat...
TRANSCRIPT
![Page 1: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/1.jpg)
A reinforcement learning model of song
acquisition in the bird
Michale Fee
McGovern Institute
Department of Brain and Cognitive Sciences
Massachusetts Institute of Technology
9.54
November 12, 2014
![Page 2: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/2.jpg)
0 kHz
10 kHz
1s
Fre
qu
en
cy
Motif Motif
Syllable
(~100ms)
Note (~10ms)
Structure of zebra finch song
![Page 3: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/3.jpg)
Songbirds learn to sing by imitating their parents
De
cre
ase
d V
aria
bili
ty
Incre
ase
d S
imila
rity to
Tu
tor
Tutor Song
Subsong
Plastic Song
Crystallized
![Page 4: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/4.jpg)
Overview
• The songbird as a model system for understanding how the brain
generates and learns complex sequential behaviors
• Review some current understanding of the mechanisms of song
production
• Describe progress in elucidating the role of cortical and basal ganglia
circuits in song learning.
• Some speculations on how insights from the songbird may inform our
understanding of mammalian BG function
![Page 5: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/5.jpg)
A circuit for vocal production
RA
nXII
Motor Pathway
HVC
Cortex
Uva
Thalamus
Nottebohm et al, 1976, 1982
![Page 6: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/6.jpg)
Hahnloser, Kozhevnikov and Fee, 2002
Antidromic activation allows identification
of RA-projecting neurons in HVC
HVC
RA
X
Stimulation
electrode
Extracellular
recording
electrode
![Page 7: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/7.jpg)
HVC neurons burst throughout the song
Lynch, Okubo and Fee, in
preparation
Leonardo and Fee (J Neurosci, 2004)
Experimental evidence in favor of a clock model and
against a gesture trajectory extrema (GTE) model of HVC encoding Galen F. Lynch1, Tatsuo Okubo1*, Michael B. Lynn1, Alexander Hanuschkin2, Richard H.R. Hahnloser2, and Michale S. Fee1
1: McGovern Institute for Brain Research, Department of Brain and Cognitive Sciences, MIT, Cambridge, MA, USA
2: Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland
LLL6
675.18
Introduction
HVC is a premotor area necessary for stereotyped adult song.
[4] Vu et al. (J. Neurosci. 1998)
[5] Long & Fee (Nature, 2008)
Songbirds learn and produce a complex sequence of vocal
gestures. •Zebra finches learn to sing by imitation
•Adults birds sing a precise, highly stereotyped song
•This learned requires precise motor control with10ms temporal resolution
• Projection neurons in HVC burst
sparsely during singing in adults [1].
• Different neurons are active at different
times
• Lesions of HVC eliminate stereotyped
elements of song [2,3].
• Disrupting HVC activity disrupts song
production [4].
• Cooling of HVC stretches stereotyped
components of the song [5]. Hahnloser et al. (Nature, 2002)
[1] Hahnloser et al. (Nature, 2002) [2] Nottebohm et al. (J Comp. Neurol., 1976) [3] Aronov et al. (Science, 2008)
Two different models of HVC coding have been proposed.
•It has been previously hypothesized that HVC neurons form a continuous
chain of activity, or ‘c
l
ock’ , that drives motor activity [6 ] .
•This model states HVC is active throughout the song, and that HVC activity
controls the timing of song.
*The Nakajima Foundation
and NIH Grant #DC009183
The clock model of HVC coding
The GTE model of HVC coding
Discriminating between models
Here we test the GTE hypotheses with a large
dataset of HVC neurons.
[6] Leonardo and Fee (J Neurosci., 2004) [7] Amador et al (Nature, 2013)
The GTE model makes testable predictions about the
statistics of HVC bursts. •The GTE model states that HVC bursts are clustered around GTE times, and
absent elsewhere [8].
•We compare the predictions of this model to a uniform null hypothesis in which
bursts are randomly placed with uniform probability.
•Clustering of bursts at GTEs makes specific predictions for
•The clustering of bursts around GTEs would be reflected in burst
statistics by an excess of short interburst intervals (IBIs). Long inter-
GTE intervals would produce gaps between bursts, and thus very long
IBIs. Combined, this would increase the variance and skew of the IBI
distriubtions compared to the variance and skew expected in the
uniform null hypothesis.
•The alignment of burst times to GTE times would be reflected by
significant cross correlation between burst times and GTE times.
HVC burst activity covers the song
motif, supporting the clock model.
Variability in burst density is
inconsistent with the GTE model
100ms
Burst #
1
66
100ms
Burst #
1
56
100ms
Burst #
1
91
Interburst intervals (IBIs) are
inconsistent with the GTE model
Bird A: 66 bursts, 40 neurons
Bird B: 56 bursts, 44 neurons
Bird C: 91 bursts, 64 neurons
Bursts are not cross correlated with
GTEs in the song
Conclusions
Bursts are not cross correlated with
interneuron minima, inconsistent with
the GTE model
20 40 60 80 100 120 0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016 P = .51
P = 4x10-4
Longest IBI [ms]
P(L
on
gest
IBI)
Distribution of longest IBIs
Density variance [bursts2/ms2]
P(D
ensi
ty v
aria
nce
)
Bird C
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0
20
40
60
80
100
120
140 P = .13
P < 1x10-6
• Amador et al. modeled the biophysics of song production as a low
dimensional system with two control parameters.
• With a relatively small dataset of projection neurons, Amador et al. observed
that burst times appeared to be aligned to extrema and discontinuities of
these control parameters, called gesture trajectory extrema (GTE).
Weak GTE hypothesis: HVC projection neurons preferentially
burst around GTEs
• Amador et al. further concluded that HVC projection neurons burst only at
GTEs, and nowhere else [7].
Strong GTE hypothesis: HVC projection neurons burst around
GTEs, and nowhere else.
• The weak GTE hypothesis is not inconsistent with the clock model.
• The strong GTE hypothesis is incompatible with the clock model, because it
predicts gaps in HVC activity.
•The GTE model predicts that HVC burst density
(#bursts per ms) has large variations
• We simulated burst density variance under the
GTE model and uniform null hypotheses
•The observed burst densities are highly
inconsistent with the GTE model and are
consistent with the null hypothesis
0
0.5
Bu
rst
De
nsi
ty
Distributions of burst density variance
0.005 0.01 0.015 0.02 0
50
100
150
200
250
300
350
Density variance [bursts2/ms2]
P(D
ensi
ty v
aria
nce
)
P = .09
P =3x10-6
GTE Null
Bird A
Distribution of the # of IBIs < 5ms
10 20 30 40 50 60 70 80 0
0.005
0.01
0.015
0.02
0.025
# IBIs
P (
#IB
Is)
P = .81
P = 0.002
GTE
Null
20 40 60 80 100 120 0
0.002
0.004
0.006
0.008
0.01
0.012
P = .68
P = 0.028
Longest IBI [ms] P
(Lo
nge
st IB
I)
Distribution of longest IBIs
0 5 10 15 20 25 30 35 40 45
0.005
0.01
0.015
0.02
0.025
0.03
Distribution of the # of IBIs < 5ms
# IBIs
P (
#IB
Is)
P = 0.011
P = .71 GTE
Null
Bird C
Bird B
Bird A
50 100 150 0
1
2
3
4
5
6
7
8
9 x 10
-3 Distribution of longest IBIs
P = 9x10-4
P = .09
Longest IBI [ms]
P(L
on
ge
st IB
I)
10 15 20 25 30 35 40 45 0
0.005
0.01
0.015
0.02
0.025
0.03
Distribution of the # of IBIs < 5ms
P = 0.003
P = .27
GTE
Null
# IBIs
P (
#IB
Is)
GTEs for bird C:
• The GTE model states that interneuron minima should preferentially occur at
GTEs.
• Thus interneuron minima should be correlated with projection neuron bursts.
• We tested this prediction by analyzing 237 minima times from 17 interneurons
and 66 bursts from 40 projection neurons recorded in a single bird
• We observe no significant correlation between minima times and burst times.
Amador et al. (Nature, 2013)
•We test the GTE hypothesis by looking for these signatures in a large dataset
of HVC projection neurons
•We determine the probability of the data under each hypothesis by generating
surrogate datasets under the uniform null hypothesis and under the GTE
hypothesis.
Long IBIs Many short IBIs
GTEs
Bursts placed under GTE model
Bursts placed under null model
50ms
50ms
Bursts:
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0
1
2
3
4
5
6
7
8
9 Distribution of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
GTE Null
P = .19
P < 3x10-6
-100 -50 0 50 100 -0.4
-0.2
0
0.2
0.4
0.6
0.8
1
offset (burst-GTE ) [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between burst and GTE times
GTE
5-95% Null 5%
FWER Bird C
-100 -50 0 50 100 -0.2
0
0.2
0.4
0.6
0.8
offset (ms)
Cro
ss c
orr
ela
tio
n
Average cross correlation between
HVC burst and GTE times for each model
GTE
Null
Transitions for bird B: 50ms
Bursts:
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0
0.5
1
1.5
2
2.5
3
3.5
Distriubtion of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
Null
P = .20
-100 -50 0 50 100 -0.4
-0.2
0
0.2
0.4
0.6
offset (burst-GTE) [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between HVC bursts and song features
Null 5%
FWER Bird A
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0
0.5
1
1.5
2
2.5
3
3.5
4 Distriubtion of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
P = .90
Null
-100 -50 0 50 100 -0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
offset bursts-minima [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between projection neuron bursts
and interneuron minima
Null 5%
FWER
Bird A
PN bursts
Int. minima
25 100
1
2
4
Variance [ms2]
Ske
w [
ms
3]
Distribution of variance vs. skew
0 10 20 30 40 50 0
10
20
30
40
IBI (ms)
0 10 20 30 40 50
IBI (ms)
Co
un
t
Interburst interval distributions
Uniform
5-95%
Bird B
GTE
5-95%
25 100 400
0.5
1
2
4
Variance [ms2]
Ske
w [
ms
3]
Distribution of variance vs. skew
25 100 400
0.5
1
2
4
Distribution of variance vs. skew
Variance [ms2]
Ske
w [
ms
3]
0 10 20 30 40 50 0
20
40
60
IBI (ms)
0 10 20 30 40 50
IBI (ms)
Co
un
t
Interburst interval distributions
Uniform
5-95%
Bird C
GTE
5-95%
0 10 20 30 40 50
IBI (ms) 0 10 20 30 40 50
0
10
20
30
40
IBI (ms)
Null
5-95%
GTE
5-95%
Co
un
t Bird A
Interburst interval distributions
•We examined various measures of the IBI distribution under the GTE and
uniform null hypotheses
•The observed IBI distributions are highly inconsistent with the GTE hypothesis
and are consistent with the null hypothesis
•The GTE model states that projection neurons should burst near GTEs.
•This predicts that GTE times would be cross correlated with bursts times.
•We are testing this by deriving GTEs from the with published techniques[7,8]. In
one bird so far, we found no correlation between burst and GTE times.
• In another bird, we found that acoustic transitions in song are not correlated with
burst times.
• We have tested the GTE model with a dataset of 213 bursts from 148
identified projection neurons in HVC, recorded from three birds.
• We find that interburst intervals are highly inconsistent with the GTE model,
and are consistent with a uniform null model.
• We additionally find that HVC burst times are not correlated with GTE times or
acoustic transitions in the song.
• We find that interneuron minima are not significantly correlated with these
projection neuron bursts.
• Our findings are highly inconsistent with the strong form of the GTE
hypothesis. In addition, our data do not reject the uniform null hypothesis, and
therefore do not support even the weak GTE hypothesis.
[8] Perl et al (Phys. Rev. E, 2011)
10 20 30 40 0
20
40
60
80
Interburst interval (IBI) [ms]
Cou
nt
Average IBI distribution for each model
GTE
Null
0.005 0.01 0.015 0.02 0.025 0.03 0.035 0
50
100
150
200
250
P = .73
P =.009
Density variance [bursts2/ms2]
P(D
ensi
ty
vari
ance
)
Bird B
Leonardo and Fee (J Neurosci, 2004)
Experimental evidence in favor of a clock model and
against a gesture trajectory extrema (GTE) model of HVC encoding Galen F. Lynch1, Tatsuo Okubo1*, Michael B. Lynn1, Alexander Hanuschkin2, Richard H.R. Hahnloser2, and Michale S. Fee1
1: McGovern Institute for Brain Research, Department of Brain and Cognitive Sciences, MIT, Cambridge, MA, USA
2: Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland
LLL6
675.18
Introduction
HVC is a premotor area necessary for stereotyped adult song.
[4] Vu et al. (J. Neurosci. 1998)
[5] Long & Fee (Nature, 2008)
Songbirds learn and produce a complex sequence of vocal
gestures. •Zebra finches learn to sing by imitation
•Adults birds sing a precise, highly stereotyped song
•This learned requires precise motor control with10ms temporal resolution
• Projection neurons in HVC burst
sparsely during singing in adults [1].
• Different neurons are active at different
times
• Lesions of HVC eliminate stereotyped
elements of song [2,3].
• Disrupting HVC activity disrupts song
production [4].
• Cooling of HVC stretches stereotyped
components of the song [5]. Hahnloser et al. (Nature, 2002)
[1] Hahnloser et al. (Nature, 2002) [2] Nottebohm et al. (J Comp. Neurol., 1976) [3] Aronov et al. (Science, 2008)
Two different models of HVC coding have been proposed.
•It has been previously hypothesized that HVC neurons form a continuous
chain of activity, or ‘c
l
ock’ , that drives motor activity [6 ] .
•This model states HVC is active throughout the song, and that HVC activity
controls the timing of song.
*The Nakajima Foundation
and NIH Grant #DC009183
The clock model of HVC coding
The GTE model of HVC coding
Discriminating between models
Here we test the GTE hypotheses with a large
dataset of HVC neurons.
[6] Leonardo and Fee (J Neurosci., 2004) [7] Amador et al (Nature, 2013)
The GTE model makes testable predictions about the
statistics of HVC bursts. •The GTE model states that HVC bursts are clustered around GTE times, and
absent elsewhere [8].
•We compare the predictions of this model to a uniform null hypothesis in which
bursts are randomly placed with uniform probability.
•Clustering of bursts at GTEs makes specific predictions for
•The clustering of bursts around GTEs would be reflected in burst
statistics by an excess of short interburst intervals (IBIs). Long inter-
GTE intervals would produce gaps between bursts, and thus very long
IBIs. Combined, this would increase the variance and skew of the IBI
distriubtions compared to the variance and skew expected in the
uniform null hypothesis.
•The alignment of burst times to GTE times would be reflected by
significant cross correlation between burst times and GTE times.
HVC burst activity covers the song
motif, supporting the clock model.
Variability in burst density is
inconsistent with the GTE model
100ms
Burst #
1
66
100ms
Burst #
1
56
100ms
Burst #
1
91
Interburst intervals (IBIs) are
inconsistent with the GTE model
Bird A: 66 bursts, 40 neurons
Bird B: 56 bursts, 44 neurons
Bird C: 91 bursts, 64 neurons
Bursts are not cross correlated with
GTEs in the song
Conclusions
Bursts are not cross correlated with
interneuron minima, inconsistent with
the GTE model
20 40 60 80 100 120 0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016 P = .51
P = 4x10-4
Longest IBI [ms]
P(L
on
gest
IBI)
Distribution of longest IBIs
Density variance [bursts2/ms2]
P(D
ensi
ty v
aria
nce
)
Bird C
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0
20
40
60
80
100
120
140 P = .13
P < 1x10-6
• Amador et al. modeled the biophysics of song production as a low
dimensional system with two control parameters.
• With a relatively small dataset of projection neurons, Amador et al. observed
that burst times appeared to be aligned to extrema and discontinuities of
these control parameters, called gesture trajectory extrema (GTE).
Weak GTE hypothesis: HVC projection neurons preferentially
burst around GTEs
• Amador et al. further concluded that HVC projection neurons burst only at
GTEs, and nowhere else [7].
Strong GTE hypothesis: HVC projection neurons burst around
GTEs, and nowhere else.
• The weak GTE hypothesis is not inconsistent with the clock model.
• The strong GTE hypothesis is incompatible with the clock model, because it
predicts gaps in HVC activity.
•The GTE model predicts that HVC burst density
(#bursts per ms) has large variations
• We simulated burst density variance under the
GTE model and uniform null hypotheses
•The observed burst densities are highly
inconsistent with the GTE model and are
consistent with the null hypothesis
0
0.5
Bu
rst
Den
sity
Distributions of burst density variance
0.005 0.01 0.015 0.02 0
50
100
150
200
250
300
350
Density variance [bursts2/ms2]
P(D
ensi
ty v
aria
nce
)
P = .09
P =3x10-6
GTE Null
Bird A
Distribution of the # of IBIs < 5ms
10 20 30 40 50 60 70 80 0
0.005
0.01
0.015
0.02
0.025
# IBIs
P (
#IB
Is)
P = .81
P = 0.002
GTE
Null
20 40 60 80 100 120 0
0.002
0.004
0.006
0.008
0.01
0.012
P = .68
P = 0.028
Longest IBI [ms]
P(L
on
gest
IBI)
Distribution of longest IBIs
0 5 10 15 20 25 30 35 40 45
0.005
0.01
0.015
0.02
0.025
0.03
Distribution of the # of IBIs < 5ms
# IBIs
P (
#IB
Is)
P = 0.011
P = .71 GTE
Null
Bird C
Bird B
Bird A
50 100 150 0
1
2
3
4
5
6
7
8
9 x 10
-3 Distribution of longest IBIs
P = 9x10-4
P = .09
Longest IBI [ms]
P(L
on
ge
st IB
I)
10 15 20 25 30 35 40 45 0
0.005
0.01
0.015
0.02
0.025
0.03
Distribution of the # of IBIs < 5ms
P = 0.003
P = .27
GTE
Null
# IBIs
P (
#IB
Is)
GTEs for bird C:
• The GTE model states that interneuron minima should preferentially occur at
GTEs.
• Thus interneuron minima should be correlated with projection neuron bursts.
• We tested this prediction by analyzing 237 minima times from 17 interneurons
and 66 bursts from 40 projection neurons recorded in a single bird
• We observe no significant correlation between minima times and burst times.
Amador et al. (Nature, 2013)
•We test the GTE hypothesis by looking for these signatures in a large dataset
of HVC projection neurons
•We determine the probability of the data under each hypothesis by generating
surrogate datasets under the uniform null hypothesis and under the GTE
hypothesis.
Long IBIs Many short IBIs
GTEs
Bursts placed under GTE model
Bursts placed under null model
50ms
50ms
Bursts:
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0
1
2
3
4
5
6
7
8
9 Distribution of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
GTE Null
P = .19
P < 3x10-6
-100 -50 0 50 100 -0.4
-0.2
0
0.2
0.4
0.6
0.8
1
offset (burst-GTE ) [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between burst and GTE times
GTE
5-95% Null 5%
FWER Bird C
-100 -50 0 50 100 -0.2
0
0.2
0.4
0.6
0.8
offset (ms)
Cro
ss c
orr
ela
tio
n
Average cross correlation between
HVC burst and GTE times for each model
GTE
Null
Transitions for bird B: 50ms
Bursts:
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0
0.5
1
1.5
2
2.5
3
3.5
Distriubtion of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
Null
P = .20
-100 -50 0 50 100 -0.4
-0.2
0
0.2
0.4
0.6
offset (burst-GTE) [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between HVC bursts and song features
Null 5%
FWER Bird A
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0
0.5
1
1.5
2
2.5
3
3.5
4 Distriubtion of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
P = .90
Null
-100 -50 0 50 100 -0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
offset bursts-minima [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between projection neuron bursts
and interneuron minima
Null 5%
FWER
Bird A
PN bursts
Int. minima
25 100
1
2
4
Variance [ms2]
Ske
w [
ms
3]
Distribution of variance vs. skew
0 10 20 30 40 50 0
10
20
30
40
IBI (ms)
0 10 20 30 40 50
IBI (ms)
Co
un
t
Interburst interval distributions
Uniform
5-95%
Bird B
GTE
5-95%
25 100 400
0.5
1
2
4
Variance [ms2]
Ske
w [
ms
3]
Distribution of variance vs. skew
25 100 400
0.5
1
2
4
Distribution of variance vs. skew
Variance [ms2]
Ske
w [
ms
3]
0 10 20 30 40 50 0
20
40
60
IBI (ms)
0 10 20 30 40 50
IBI (ms)
Co
un
t
Interburst interval distributions
Uniform
5-95%
Bird C
GTE
5-95%
0 10 20 30 40 50
IBI (ms) 0 10 20 30 40 50
0
10
20
30
40
IBI (ms)
Null
5-95%
GTE
5-95%
Co
un
t Bird A
Interburst interval distributions
•We examined various measures of the IBI distribution under the GTE and
uniform null hypotheses
•The observed IBI distributions are highly inconsistent with the GTE hypothesis
and are consistent with the null hypothesis
•The GTE model states that projection neurons should burst near GTEs.
•This predicts that GTE times would be cross correlated with bursts times.
•We are testing this by deriving GTEs from the with published techniques[7,8]. In
one bird so far, we found no correlation between burst and GTE times.
• In another bird, we found that acoustic transitions in song are not correlated with
burst times.
• We have tested the GTE model with a dataset of 213 bursts from 148
identified projection neurons in HVC, recorded from three birds.
• We find that interburst intervals are highly inconsistent with the GTE model,
and are consistent with a uniform null model.
• We additionally find that HVC burst times are not correlated with GTE times or
acoustic transitions in the song.
• We find that interneuron minima are not significantly correlated with these
projection neuron bursts.
• Our findings are highly inconsistent with the strong form of the GTE
hypothesis. In addition, our data do not reject the uniform null hypothesis, and
therefore do not support even the weak GTE hypothesis.
[8] Perl et al (Phys. Rev. E, 2011)
10 20 30 40 0
20
40
60
80
Interburst interval (IBI) [ms]
Co
un
t
Average IBI distribution for each model
GTE
Null
0.005 0.01 0.015 0.02 0.025 0.03 0.035 0
50
100
150
200
250
P = .73
P =.009
Density variance [bursts2/ms2]
P(D
ensi
ty
vari
ance
)
Bird B
Leonardo and Fee (J Neurosci, 2004)
Experimental evidence in favor of a clock model and
against a gesture trajectory extrema (GTE) model of HVC encoding Galen F. Lynch1, Tatsuo Okubo1*, Michael B. Lynn1, Alexander Hanuschkin2, Richard H.R. Hahnloser2, and Michale S. Fee1
1: McGovern Institute for Brain Research, Department of Brain and Cognitive Sciences, MIT, Cambridge, MA, USA
2: Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland
LLL6
675.18
Introduction
HVC is a premotor area necessary for stereotyped adult song.
[4] Vu et al. (J. Neurosci. 1998)
[5] Long & Fee (Nature, 2008)
Songbirds learn and produce a complex sequence of vocal
gestures. •Zebra finches learn to sing by imitation
•Adults birds sing a precise, highly stereotyped song
•This learned requires precise motor control with10ms temporal resolution
• Projection neurons in HVC burst
sparsely during singing in adults [1].
• Different neurons are active at different
times
• Lesions of HVC eliminate stereotyped
elements of song [2,3].
• Disrupting HVC activity disrupts song
production [4].
• Cooling of HVC stretches stereotyped
components of the song [5]. Hahnloser et al. (Nature, 2002)
[1] Hahnloser et al. (Nature, 2002) [2] Nottebohm et al. (J Comp. Neurol., 1976) [3] Aronov et al. (Science, 2008)
Two different models of HVC coding have been proposed.
•It has been previously hypothesized that HVC neurons form a continuous
chain of activity, or ‘c
l
ock’ , that drives motor activity [6 ] .
•This model states HVC is active throughout the song, and that HVC activity
controls the timing of song.
*The Nakajima Foundation
and NIH Grant #DC009183
The clock model of HVC coding
The GTE model of HVC coding
Discriminating between models
Here we test the GTE hypotheses with a large
dataset of HVC neurons.
[6] Leonardo and Fee (J Neurosci., 2004) [7] Amador et al (Nature, 2013)
The GTE model makes testable predictions about the
statistics of HVC bursts. •The GTE model states that HVC bursts are clustered around GTE times, and
absent elsewhere [8].
•We compare the predictions of this model to a uniform null hypothesis in which
bursts are randomly placed with uniform probability.
•Clustering of bursts at GTEs makes specific predictions for
•The clustering of bursts around GTEs would be reflected in burst
statistics by an excess of short interburst intervals (IBIs). Long inter-
GTE intervals would produce gaps between bursts, and thus very long
IBIs. Combined, this would increase the variance and skew of the IBI
distriubtions compared to the variance and skew expected in the
uniform null hypothesis.
•The alignment of burst times to GTE times would be reflected by
significant cross correlation between burst times and GTE times.
HVC burst activity covers the song
motif, supporting the clock model.
Variability in burst density is
inconsistent with the GTE model
100ms
Burst #
1
66
100ms
Burst #
1
56
100ms
Burst #
1
91
Interburst intervals (IBIs) are
inconsistent with the GTE model
Bird A: 66 bursts, 40 neurons
Bird B: 56 bursts, 44 neurons
Bird C: 91 bursts, 64 neurons
Bursts are not cross correlated with
GTEs in the song
Conclusions
Bursts are not cross correlated with
interneuron minima, inconsistent with
the GTE model
20 40 60 80 100 120 0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016 P = .51
P = 4x10-4
Longest IBI [ms]
P(L
on
gest
IBI)
Distribution of longest IBIs
Density variance [bursts2/ms2]
P(D
ensi
ty v
aria
nce
)
Bird C
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0
20
40
60
80
100
120
140 P = .13
P < 1x10-6
• Amador et al. modeled the biophysics of song production as a low
dimensional system with two control parameters.
• With a relatively small dataset of projection neurons, Amador et al. observed
that burst times appeared to be aligned to extrema and discontinuities of
these control parameters, called gesture trajectory extrema (GTE).
Weak GTE hypothesis: HVC projection neurons preferentially
burst around GTEs
• Amador et al. further concluded that HVC projection neurons burst only at
GTEs, and nowhere else [7].
Strong GTE hypothesis: HVC projection neurons burst around
GTEs, and nowhere else.
• The weak GTE hypothesis is not inconsistent with the clock model.
• The strong GTE hypothesis is incompatible with the clock model, because it
predicts gaps in HVC activity.
•The GTE model predicts that HVC burst density
(#bursts per ms) has large variations
• We simulated burst density variance under the
GTE model and uniform null hypotheses
•The observed burst densities are highly
inconsistent with the GTE model and are
consistent with the null hypothesis
0
0.5
Bu
rst
Den
sity
Distributions of burst density variance
0.005 0.01 0.015 0.02 0
50
100
150
200
250
300
350
Density variance [bursts2/ms2]
P(D
ensi
ty v
aria
nce
)
P = .09
P =3x10-6
GTE Null
Bird A
Distribution of the # of IBIs < 5ms
10 20 30 40 50 60 70 80 0
0.005
0.01
0.015
0.02
0.025
# IBIs
P (
#IB
Is)
P = .81
P = 0.002
GTE
Null
20 40 60 80 100 120 0
0.002
0.004
0.006
0.008
0.01
0.012
P = .68
P = 0.028
Longest IBI [ms]
P(L
on
gest
IBI)
Distribution of longest IBIs
0 5 10 15 20 25 30 35 40 45
0.005
0.01
0.015
0.02
0.025
0.03
Distribution of the # of IBIs < 5ms
# IBIs
P (
#IB
Is)
P = 0.011
P = .71 GTE
Null
Bird C
Bird B
Bird A
50 100 150 0
1
2
3
4
5
6
7
8
9 x 10
-3 Distribution of longest IBIs
P = 9x10-4
P = .09
Longest IBI [ms]
P(L
on
ge
st IB
I)
10 15 20 25 30 35 40 45 0
0.005
0.01
0.015
0.02
0.025
0.03
Distribution of the # of IBIs < 5ms
P = 0.003
P = .27
GTE
Null
# IBIs
P (
#IB
Is)
GTEs for bird C:
• The GTE model states that interneuron minima should preferentially occur at
GTEs.
• Thus interneuron minima should be correlated with projection neuron bursts.
• We tested this prediction by analyzing 237 minima times from 17 interneurons
and 66 bursts from 40 projection neurons recorded in a single bird
• We observe no significant correlation between minima times and burst times.
Amador et al. (Nature, 2013)
•We test the GTE hypothesis by looking for these signatures in a large dataset
of HVC projection neurons
•We determine the probability of the data under each hypothesis by generating
surrogate datasets under the uniform null hypothesis and under the GTE
hypothesis.
Long IBIs Many short IBIs
GTEs
Bursts placed under GTE model
Bursts placed under null model
50ms
50ms
Bursts:
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0
1
2
3
4
5
6
7
8
9 Distribution of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
GTE Null
P = .19
P < 3x10-6
-100 -50 0 50 100 -0.4
-0.2
0
0.2
0.4
0.6
0.8
1
offset (burst-GTE ) [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between burst and GTE times
GTE
5-95% Null 5%
FWER Bird C
-100 -50 0 50 100 -0.2
0
0.2
0.4
0.6
0.8
offset (ms)
Cro
ss c
orr
ela
tio
n
Average cross correlation between
HVC burst and GTE times for each model
GTE
Null
Transitions for bird B: 50ms
Bursts:
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0
0.5
1
1.5
2
2.5
3
3.5
Distriubtion of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
Null
P = .20
-100 -50 0 50 100 -0.4
-0.2
0
0.2
0.4
0.6
offset (burst-GTE) [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between HVC bursts and song features
Null 5%
FWER Bird A
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0
0.5
1
1.5
2
2.5
3
3.5
4 Distriubtion of cross correlations at 0ms delay
Cross Correlation at 0ms
P(C
orr
ela
tio
n)
P = .90
Null
-100 -50 0 50 100 -0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
offset bursts-minima [ms]
Cro
ss c
orr
ela
tio
n
Cross correlation between projection neuron bursts
and interneuron minima
Null 5%
FWER
Bird A
PN bursts
Int. minima
25 100
1
2
4
Variance [ms2]
Ske
w [
ms
3]
Distribution of variance vs. skew
0 10 20 30 40 50 0
10
20
30
40
IBI (ms)
0 10 20 30 40 50
IBI (ms)
Co
un
t
Interburst interval distributions
Uniform
5-95%
Bird B
GTE
5-95%
25 100 400
0.5
1
2
4
Variance [ms2]
Ske
w [
ms
3]
Distribution of variance vs. skew
25 100 400
0.5
1
2
4
Distribution of variance vs. skew
Variance [ms2]
Ske
w [
ms
3]
0 10 20 30 40 50 0
20
40
60
IBI (ms)
0 10 20 30 40 50
IBI (ms)
Co
un
t
Interburst interval distributions
Uniform
5-95%
Bird C
GTE
5-95%
0 10 20 30 40 50
IBI (ms) 0 10 20 30 40 50
0
10
20
30
40
IBI (ms)
Null
5-95%
GTE
5-95%
Co
un
t Bird A
Interburst interval distributions
•We examined various measures of the IBI distribution under the GTE and
uniform null hypotheses
•The observed IBI distributions are highly inconsistent with the GTE hypothesis
and are consistent with the null hypothesis
•The GTE model states that projection neurons should burst near GTEs.
•This predicts that GTE times would be cross correlated with bursts times.
•We are testing this by deriving GTEs from the with published techniques[7,8]. In
one bird so far, we found no correlation between burst and GTE times.
• In another bird, we found that acoustic transitions in song are not correlated with
burst times.
• We have tested the GTE model with a dataset of 213 bursts from 148
identified projection neurons in HVC, recorded from three birds.
• We find that interburst intervals are highly inconsistent with the GTE model,
and are consistent with a uniform null model.
• We additionally find that HVC burst times are not correlated with GTE times or
acoustic transitions in the song.
• We find that interneuron minima are not significantly correlated with these
projection neuron bursts.
• Our findings are highly inconsistent with the strong form of the GTE
hypothesis. In addition, our data do not reject the uniform null hypothesis, and
therefore do not support even the weak GTE hypothesis.
[8] Perl et al (Phys. Rev. E, 2011)
10 20 30 40 0
20
40
60
80
Interburst interval (IBI) [ms]
Cou
nt
Average IBI distribution for each model
GTE
Null
0.005 0.01 0.015 0.02 0.025 0.03 0.035 0
50
100
150
200
250
P = .73
P =.009
Density variance [bursts2/ms2]
P(D
ensi
ty
vari
ance
)
Bird B
![Page 8: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/8.jpg)
Activity of RA neurons during
singing
Yu and Margoliash, 1996
Motif
Leonardo and Fee, 2005
HVC
RA
Extracellular
recording
electrode
![Page 9: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/9.jpg)
Simple sequence generation circuit
Sparse representation of time
Output
Leonardo and Fee, 2005
![Page 10: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/10.jpg)
Sparse representation of time
Output
Simple sequence generation circuit
Leonardo and Fee, 2005
![Page 11: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/11.jpg)
HVC is the ‘clock’ of the song motor
pathway
Brain cooling to localize dynamics
np
RA
nXII
HVC0.0A
0.25A
-0.25A
-0.5A
-0.75A
...
Bilateral cooling of HVC causes
uniform slowing of the song
5 mm
Long and Fee, Nature 2008
![Page 12: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/12.jpg)
A simple reinforcement model of song
learning
Song motor
systemSong Exploratory
variability
Song evaluation
Auditory
feedback
Auditory Memory
Doya and Sejnowski, 1989
--
Error/Reinforcement signal
![Page 13: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/13.jpg)
RA
nXII
Motor Pathway
HVC
A separate circuit for song learning
Cortex
LMAN
DLM
Thalamus
Anterior Forebrain Pathway
(AFP)
•The learning pathway is not necessary for adult song production , but is required
for learning (Bottjer, 1984, Scharff and Nottebohm, 1991)
Instructive signal
•Bottjer proposed that the AFP transmits an instructive signal that guides plasticity
in the motor pathway
Basal Ganglia
Area X
![Page 14: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/14.jpg)
RA
nXII
LMAN
HVC
Separate premotor pathways for
stereotyped song and variability
Sequence generator
Variability generator
Kao et al, 2005
Ölveczky et al, 2005
Aronov et al, 2008
Stepanek and Doupe, 2010
![Page 15: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/15.jpg)
RA
nXII
LMAN
HVC
TTX or Muscimol
Separate premotor pathways for
stereotyped song and variability
Sequence generator
Variability generator
Kao et al, 2005
Ölveczky et al, 2005
Aronov et al, 2008
Stepanek and Doupe, 2010
![Page 16: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/16.jpg)
Transient inactivation of the learning pathway
55 day old bird
RA
nXII
LMAN
HVC
Olveczky, Andalman, and Fee, 2005
![Page 17: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/17.jpg)
LMAN drives exploratory variability in song
LMAN intact
LMAN inactivated
0 20 40 60
-20
0
20
Resid
ua
l P
itch (
Hz)
Time (ms)
![Page 18: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/18.jpg)
LMAN intact
250 ms
LMAN inactivated
30 dB
LMAN also drives early song ‘babbling’
Goldberg and Fee, 2011
![Page 19: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/19.jpg)
RA
nXII
LMAN
Motor Pathway
Learning Pathway (AFP)
HVC
HVC lesions abolish all stereotyped song structure
![Page 20: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/20.jpg)
HVC lesions abolish all stereotyped song structure
Pre HVC lesion Post HVC lesion
Transient pharmacological inactivation of HVC produces the same effect
Aronov, Andalman and Fee, Science 2008,
Subsong bird
Plastic song bird
Adult bird
![Page 21: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/21.jpg)
The basal ganglia are not necessary for
subsong or vocal variability in juvenile birds
30
dB
SubsongPre-lesion
250 ms
Post-lesion
HVC
RA
nXIIts
LMAN
XDLM X
Goldberg and Fee, 2011
• Lesions of the BG have little or no acute effect on juvenile song variability.
• Local cooling in LMAN slow timescales of babbling exploratory vocal variability is generated by local circuit dynamics within LMAN.
![Page 22: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/22.jpg)
RA
nXII
Separate premotor pathways for
stereotyped song and variability
HVC
Sequence generator
LMAN
Variability generator
Kao et al, 2005
Ölveczky et al, 2005
Aronov et al, 2008
Stepanek and Doupe, 2010
![Page 23: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/23.jpg)
RA
nXII
LMAN
HVC
Separate premotor pathways for
stereotyped song and variability
Sequence generator
Variability generator
Kao et al, 2005
Ölveczky et al, 2005
Aronov et al, 2008
Stepanek and Doupe, 2010
Instructive signal
![Page 24: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/24.jpg)
RA
nXII
LMAN
HVC
Separate premotor pathways for
stereotyped song and variability
Sequence generator
Variability generator
Instructive signal
Area XDLM
![Page 25: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/25.jpg)
Tchernichovski, Mitra, Lints, Nottebohm, 2001
TutorPupil
Days of Training
5 8 12 20 30
606 Hz
Harmonic
Stack
Pitch (Hz) 568 554 551 596 607
Song learning is slow
![Page 26: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/26.jpg)
Experimental control of song learning
0.5
0.6
Targeted region
of syllable
Vocaliz
ed
Pitch
(kH
z)
Heard
Feedback Noise
Pitch threshold
Andalman and Fee 2009; Tumer and Brainard 2007
Speaker
DSP
Brain
Cranial airsac
Mic
![Page 27: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/27.jpg)
Conditional auditory feedback drives pitch
learning
Tumer and Brainard 2007
Pitch (
Hz)
550
650
25 ms
0 h 2 h 4 h
Pitch (
Hz)
550
650
Andalman and Fee 2009
![Page 28: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/28.jpg)
-50 0 500
5
10
15
ΔPitch, Day (Hz)
Ob
se
rva
tio
ns
Many days of sequential learning
120 125 130 135 140 145 150 155 160 165
500
600
700
141 142 143 144470
600
161 162 163 164520
650
Pitch
(H
z)
Days Post Hatch Days Post Hatch
Days Post Hatch
Pitch
(H
z)
-50 0 500
5
10
15
ΔPitch, Overnight (Hz)
Ob
se
rva
tio
ns Up Days
Down Days
Up Days
Down Days
![Page 29: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/29.jpg)
Motor parameter space
AFP-driven
variability
Where does this learning occur in the song
control circuit?
Motor pathway
![Page 30: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/30.jpg)
Motor parameter space
AFP-driven
variability
AFP-driven
bias
Error gradient
(reduced error)
Motor pathway
Where does this learning occur in the song
control circuit?
![Page 31: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/31.jpg)
Motor parameter space
AFP-driven
variability
AFP-driven
bias
Error gradient
(reduced error)
Motor pathway
Plasticity
in motor
pathway
Where does this learning occur in the song
control circuit?
![Page 32: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/32.jpg)
AFP-driven
bias
Plasticity
in motor
pathway
HVC
RA
nXIIts
LMAN
XDLM
Motor PathwayAnterior Forebrain Pathway (AFP)
HVC
RA
nXIIts
LMAN
XDLM
Where does this learning occur in the song
control circuit?
![Page 33: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/33.jpg)
during LMAN inactivation
25 Hz
Pitch
(H
z) TTX
2 h ∆Pitch (Hz)
-50 0 500
2
4
6
8 TTX
Ob
se
rva
tio
ns
Up Days
Down Days
Δ
Vehicle
Pitch
(H
z)
Vehicle
∆Pitch (Hz)-50 0 50
0
2
4
6
8
10
Up Days
Down Days
Ob
se
rva
tio
ns
Andalman and Fee, PNAS 2009
470
600
TTX TTX TTX TTXPBS PBS PBS PBS
?
Pitch
(H
z)
drug
reservoir
cap
inflow tube
outlet tube
dialysis membrane
skull
dental acrylic
LMAN
![Page 34: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/34.jpg)
Does AFP-driven variability become biased to
reduce vocal errors?
Motor parameter space
AFP-driven
variability
AFP-driven
error-reducing
bias
Error gradient
(reduced error)
Plasticity
in motor
pathway
Yes!!
Motor pathway
![Page 35: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/35.jpg)
Is all song learning mediated by AFP bias?
120 125 130 135 140 145 150 155 160 165
500
600
700
Days Post Hatch
Pitch
(H
z)
Many days of sequential learning
![Page 36: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/36.jpg)
120 125 130 135 140 145 150 155 160 165
500
600
Days post-hatch
Pitch (
Hz)
baseline
LMAN(+)
LMAN(-)
Is all song learning mediated by AFP bias?
120 125 130 135 140 145 150 155 160 165
500
600
Days post-hatch
Pitch (
Hz)
baseline
LMAN(+)
LMAN(-)
![Page 37: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/37.jpg)
AFP bias is highly predictive of motor pathway plasticity
within the next 24 hoursP
itch
β
Day 1
Δm
Nig
ht
Day 2 Day 3
Andalman and Fee, 2009
Warren et al, 2011
-4 -2 0 2 40
0.2
0.4
0.6
0.8
1
Lag (days)
Corr
ela
tion C
oeffic
ient (r
2)
-100 0 1000
50
100
Lag = -2 d Lag = -1 d Lag = 0 d
Δm
(H
z)
(dow
n d
ays in
v.)
Estimated AFP bias (Hz, down days inverted)
Days
![Page 38: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/38.jpg)
Motor parameter space
AFP-driven
variability
AFP-driven
bias
motor
pathway
plasticityMotor pathway
motor
pathway
plasticity
Error gradient
(reduced error)
Day 1 Day 2 Day 3
Motor pathway plasticity appears to ‘integrate’AFP bias
![Page 39: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/39.jpg)
X
How is AFP bias generated?
RA
nXII
HVC
LMAN
• Area X receives an efference copy of variability signals sent to RA.
• If Area X also receives an evaluation signal, then X could figure out which variations
lead to better song performance.
• Dopaminergic midbrain (VTA) has been shown to signal reward prediction error
• Do X-projecting VTA neurons carry error-related signals?
VTA
Schultz, 2000
![Page 40: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/40.jpg)
A descending pathway from higher-order auditory areas
to VTA/SNc
AIV
RA
VTA
X
CM
Aud
Keller and Hahnloser, 2008
Gale, Perkel 2008
Mandelblat-Cerf et al, 2014
Retrograde label from VTA
AIV
Ventral Intermediate Arcopallium (AIV)
![Page 41: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/41.jpg)
X
Is AIV necessary for song learning?
nXII
HVC
LMAN
VTA
NeuN
Las, Denisenko, Mandelblat-Cerf, eLife, 2014
![Page 42: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/42.jpg)
Is AIV necessary for song learning?
Bird tutored in
home cage
AIV lesion
40 90
Check
imitationBird isolated
Age (days post hatch)
![Page 43: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/43.jpg)
AIV lesioned pupil #2 – Adult song
Example 1
Example 2
Tutor
AIV lesion produces profound song
learning deficits
![Page 44: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/44.jpg)
AIV lesions produce profound song
learning deficits
AIV lesioned –
adult song
Tutor
Lesioned controlUnlesioned controlAIV lesion
Similarity of
unrelated birds
![Page 45: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/45.jpg)
Do AIV neurons transmit an ‘error’ signal to
VTA during singing?
XnXII
HVC
LMAN
VTA-2 0 2 4 6 8 10 12 14
-0.4
-0.2
0.0
0.2
0.4
Vo
lta
ge
(m
V)
Time from stim (ms)
![Page 46: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/46.jpg)
200 ms
Noise burst
Do AIV neurons transmit an ‘error’ signal to
VTA during singing?
![Page 47: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/47.jpg)
AIV neurons show error-related signals
Mandelblat-Cerf, Las, Denisenko, under review
![Page 48: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/48.jpg)
A descending pathway from higher-order auditory areas
to VTA/SNc
AIV
RA
VTA
X
CM
Aud
Keller and Hahnloser, 2008
Gale, Perkel 2008
Mandelblat-Cerf et al, 2014
Retrograde label from VTA
AIV
Ventral Intermediate Arcopallium (AIV)
![Page 49: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/49.jpg)
RA
nXII
HVC
LMAN
How it all works: a hypothesis
![Page 50: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/50.jpg)
VTA
X
How it all works: a hypothesis
CM
Aud
![Page 51: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/51.jpg)
X
VTA
LMAN
HVC
How it all works: a hypothesis
![Page 52: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/52.jpg)
DLM
X
LMAN
HVC
How it all works: a hypothesis
![Page 53: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/53.jpg)
RA
HVC
LMAN
How it all works: a hypothesis
![Page 54: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/54.jpg)
321
VTA
HVC (Time Sequence)
LMAN
Pallidal
Thalamus
MSN
To RA
Area X
HVC(X) firing patterns
4 k
Hz
1
2
3
4
5
6
7
100 ms
A model of basal ganglia function with
functionally distinct inputs for context, motor
efference copy, and reward
The AFP forms a classic
cortical-BG-thalamo-cortical loop
![Page 55: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/55.jpg)
321
VTA
HVC (Time Sequence)
LMAN
Pallidal
Thalamus
MSN
To RA
Area X
Learning rule:
Strengthen HVC synapse
after coincidence
of LMAN, HVC and DA inputs VTA
A model of basal ganglia function with
functionally distinct inputs for context, motor
efference copy, and reward
![Page 56: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/56.jpg)
321
VTA
HVC (Time Sequence)
LMAN
Pallidal
Thalamus
MSN
To RA
Area X
A model of basal ganglia function with
functionally distinct inputs for context, motor
efference copy, and reward
Time-dependent
bias of one LMAN
neuron Goldberg and Fee 2010
LMAN
HVC
1
2
3
MSN
To RA
![Page 57: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/57.jpg)
321
VTA
HVC (Time Sequence)
LMAN
Pallidal
Thalamus
MSN
To RA
Area X
HVC synapses
Timing
Drive MSNs
Plastic
Selective for single synapses
LMAN synapses
Action
Do NOT drive MSNs
Not plastic
Global signal
A model of basal ganglia function with
functionally distinct inputs for context, motor
efference copy, and reward
![Page 58: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/58.jpg)
VTALMAN
MSN
HVC
LMAN
MSN
HVC
A learning rule with an eligibility trace allows
delayed reward
LMAN
HVC
VTA
ΔWHVC-X
Eligibility
traceEHVC-X
EHVC-X = LH
DWHVC-X = bEHVC-XR
![Page 59: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/59.jpg)
HVC 1 HVC 2
LMAN
MSN
Hypothesis for HVC-LMAN synaptic interaction
on striatal MSNs
HVC on spines
LMAN on
dendritic shafts
![Page 60: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/60.jpg)
t = 1
HVC 1 HVC 2
LMAN
MSN
Hypothesis for HVC-LMAN synaptic interaction
on striatal MSNs
![Page 61: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/61.jpg)
t = 2
HVC 1 HVC 2
LMAN
MSN
Hypothesis for HVC-LMAN synaptic interaction
on striatal MSNs
![Page 62: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/62.jpg)
t = 2
HVC 1 HVC 2
LMAN
MSN
Hypothesis for HVC-LMAN synaptic interaction
on striatal MSNs
![Page 63: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/63.jpg)
t = 2
HVC 1 HVC 2
LMAN
MSN
Hypothesis for HVC-LMAN synaptic interaction
on striatal MSNs
![Page 64: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/64.jpg)
t = 2
HVC 1 HVC 2
LMAN
MSN
Hypothesis for HVC-LMAN synaptic interaction
on striatal MSNs
Dopamine
![Page 65: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/65.jpg)
HVC 1 HVC 2
LMAN
MSN
Synapse
Strengthened
Hypothesis for HVC-LMAN synaptic interaction
on striatal MSNs
![Page 66: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/66.jpg)
Serial Block Face Scanning EM
Collaboration with Winfried Denk and Jörgen Kornfeld
![Page 67: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/67.jpg)
LMAN
200 µm
200 µm
HVC
200 µm
200 µm
Distinct morphology of HVC and LMAN axons
Michael Stetner
Axonal arbor of LMAN neuron in Area X Axonal arbor of HVC neuron in Area X
![Page 68: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/68.jpg)
MSN HVC-like LMAN-like
~94% of synapses onto spines are
from HVC-like axons
Inputs onto MSN spines originate primarily from HVC
Putative LMAN axons
Putative HVC axons
![Page 69: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/69.jpg)
The role of the basal ganglia in songbird
vocal learning
• LMAN directly drives ‘exploratory variability’ in the song motor pathway.
• LMAN-driven variability becomes biased during learning, in the direction of
improved song performance.
• We have found evidence that a dopaminergic pathway to the songbird BG may
carry ‘performance’ error-related information.
• We hypothesize that the basal ganglia determine which song variations lead to
better performance and bias the variability in the direction of improved
performance.
• We have proposed a testable model of basal ganglia function that explicitly
incorporates an efference copy of cortically-generated motor actions.
![Page 70: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/70.jpg)
The Fee Lab
Current Lab Members
• Anusha Narayan
• Natalia Denissenko
• Tatsuo Okubo
• Michael Stetner
• Emily Mackevicius
• Galen Lynch
web.mit.edu/feelab
Former Lab Members
• Richard Hahnloser
• Alexay Kozhevnikov
• Anthony Leonardo
• Michael Long
• Bence Ölveczky
• Dmitriy Aronov
• Aaron Andalman
• Lena Veit
• Jakob Förster
• Liora Las
• Jesse Goldberg
• Yael MandelblatFunding: National Institutes of Health - NIMH, NIDCD
![Page 71: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/71.jpg)
Separate premotor pathways for
stereotyped song and variability
RA
Sequence
Stereotypy
Precision
HVC
UvaRandomness
Variability
Exploration
LMAN
DLM
Motor Output
SubsongAdult song X
![Page 72: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/72.jpg)
LMAN drives subsong
Stimulating electrode
nXII
LMAN
Area XDLM
HVC
RA
![Page 73: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/73.jpg)
Insta
nta
ne
ous
firing r
ate
(Hz)
700
0
20 dB
Sound amplitude
250 ms
LMAN(RA) neurons exhibit premotor correlation
with subsong syllables
![Page 74: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/74.jpg)
Insta
nta
ne
ous
firing r
ate
(Hz)
700
0
20 dB
Sound amplitude
200 ms
LMAN(RA) neurons exhibit premotor correlation
with subsong syllables
![Page 75: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/75.jpg)
20 dB
700
0
Insta
nta
neous
firing r
ate
(Hz)
Neuron 14
Sound amplitude
250 ms
*
*
-100 0 100 2000
20
40
60
Time relative to offset (ms)
Mean f
irin
g r
ate
(H
z)
Sound amplitude
LMAN(RA) neurons exhibit premotor correlation
with subsong acoustic structure
![Page 76: A reinforcement learning model of song acquisition in the bird9.54/fall14/slides/Class19_Fee.pdfthat burst times app eared to b e aligne d to ex trema and disco ntinu ities of these](https://reader034.vdocument.in/reader034/viewer/2022042122/5e9c3773a4478b4a035d2c60/html5/thumbnails/76.jpg)
Summary
• The AFP can generate a direct premotor bias that reduces vocal errors.
• The learning accumulated across many days of training is encoded primarily in plasticity in the motor pathway.
• The contribution of the AFP is limited to the learning that occurred most recently (during the same day).
• AFP bias is predictive of subsequent plasticity in the motor pathway within the next 24 hours