NEURAL NETWORKS Biological analogy Introduction to Artificial Neural Networks Typical architectures.
Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural...
Transcript of Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural...
![Page 1: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/1.jpg)
Biological Modeling
of Neural Networks:
Week 13 – Membrane potential
densities and Fokker-Planck
Wulfram Gerstner
EPFL, Lausanne, Switzerland
13.1 Review: Integrate-and-fire
- stochastic spike arrival
13.2 Density of membrane potential - Continuity equation
13.3 Flux - jump flux
- drift flux
13.4. Fokker –Planck Equation
- free solution
13.5. Threshold and reset - time dependent activity
- network states
Week 13 – Membrane Potential Densities and Fokker-Planck Equation
![Page 2: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/2.jpg)
Spike reception
iu
i j
Spike emission
-spikes are events
-threshold
-spike/reset/refractoriness
Week 13-part 1: Review: integrate-and-fire-type models
![Page 3: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/3.jpg)
)()( tRIuuudt
deq
)();( equuVtRIVVdt
d
iui
I
j
Week 13-part 1: Review: leaky integrate-and-fire model
resetuu If firing:
![Page 4: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/4.jpg)
)()( tRIuuudt
deq LIF
resetuu If firing:
I=0 u
dt
d
u
I>0 u
dt
d
u
resting
t
u
repetitive
t
flow (drift)
towards
threshold
Week 13-part 1: Review: leaky integrate-and-fire model
![Page 5: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/5.jpg)
I(t)
)(tAn
Week 13-part 1: Review: microscopic vs. macroscopic
![Page 6: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/6.jpg)
I(t)
?
t
t
tN
tttntA
);()(population
activity
Homogeneous
network: -each neuron receives input
from k neurons in network
-each neuron receives the same
(mean) external input
Week 13-part 1: Review: homogeneous population
![Page 7: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/7.jpg)
Stochastic spike arrival:
excitation, total rate Re
inhibition, total rate Ri
u
0u
EPSC IPSC
Synaptic current pulses
)()()( ttIRuuudt
d meaneq
Langevin equation,
Ornstein Uhlenbeck process
)()()(
','
''
,
fk
fki
fk
fkeeq ttqttqRuuu
dt
d
Week 13-part 1: Review: diffusive noise/stochastic spike arrival
![Page 8: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/8.jpg)
Biological Modeling
of Neural Networks:
Week 13 – Membrane potential
densities and Fokker-Planck
Wulfram Gerstner
EPFL, Lausanne, Switzerland
13.1 Review: Integrate-and-fire
- stochastic spike arrival
13.2 Density of membrane potential -
13.3 Flux -
-
13.4. Fokker –Planck Equation
- -
13.5. Threshold and reset -
Week 13 part 2 – Membrane Potential Densities
![Page 9: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/9.jpg)
EPSC IPSC
For any arbitrary neuron in the population
Blackboard:
density of potentials?
))()((
','
''
,
fk
fki
fk
fke ttqttqRuu
dt
d
,
( ) ( )f extek
k f
qd uu t t I t
dt C
external current
input excitatory input spikes
Week 13-part 2: membrane potential density
![Page 10: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/10.jpg)
( , ) ( , )d d
p u t J u tdt du
Week 13-part 2: continuity equation
![Page 11: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/11.jpg)
u
p(u)
Exercise 1: flux caused by stochastic spike arrival
u
Membrane potential density
spike arrival rate
Next lecture:
10h15
b)
c) k
k
spike arrival rate
a) Jump at time t
Reference level u0
What is the flux
across u0?
( ) ( )}f
eq e
f
du u u R q t t
dt
![Page 12: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/12.jpg)
Biological Modeling
of Neural Networks:
Week 13 – Membrane potential
densities and Fokker-Planck
Wulfram Gerstner
EPFL, Lausanne, Switzerland
13.1 Review: Integrate-and-fire
- stochastic spike arrival
13.2 Density of membrane potential - continuity equation
-
13.3 Flux - jump flux
- drift flux
13.4. Fokker –Planck Equation
-
13.5. Threshold and reset -
Week 13 – Membrane Potential Densities and Fokker-Planck Equation
![Page 13: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/13.jpg)
Week 13-part 2: membrane potential density: flux by jumps
![Page 14: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/14.jpg)
Week 13-part 2: membrane potential density: flux by drift
![Page 15: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/15.jpg)
u
p(u)
flux – two possibilities
u
Membrane potential density
spike arrival rate a) Jumps caused at
Reference level u0
What is the flux
across u0?
b)
flux caused by jumps due to
stochastic spike arrival
flux caused by
systematic drift
Blackboard:
Slope and
density of potentials
( )( ) ( )}ext t f
eq e
f
du u u R I q t t
dt
![Page 16: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/16.jpg)
Biological Modeling
of Neural Networks:
Week 13 – Membrane potential
densities and Fokker-Planck
Wulfram Gerstner
EPFL, Lausanne, Switzerland
13.1 Review: Integrate-and-fire
- stochastic spike arrival
13.2 Density of membrane potential -
13.3 Flux - continuity equation
13.4. Fokker –Planck Equation
- source and sink
-
13.5. Threshold and reset -
Week 13 – Membrane Potential Densities and Fokker-Planck Equation
![Page 17: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/17.jpg)
EPSC IPSC
For any arbitrary neuron in the population
'
'
, ', '
1( ) ( ) ( )f f exte i
k k
k f k f
q qd uu t t t t I t
dt C C C
external input
( , ) ( , )p u t J u tt u
Continuity equation:
Flux: - jump (spike arrival)
- drift (slope of trajectory)
Blackboard:
Derive Fokker-Planck
equation
Week 13-part 4: from continuity equation to Fokker-Planck
![Page 18: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/18.jpg)
22
2( , ) [ ( ) ( , )] ( , )p u t u p u t p u t
t u u
Fokker-Planck
drift diffusion
k
kk wuu )( 2 21
2k k
k
w
spike arrival rate
Membrane potential density
u
p(u)
Week 13-part 4: Fokker-Planck equation
![Page 19: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/19.jpg)
u
p(u)
Exercise 2: solution of free Fokker-Planck equation
u
Membrane potential density: Gaussian
22
2( , ) [ ( ) ( , )] ( , )p u t u p u t p u t
t u u
Fokker-Planck
drift diffusion ( ) ( )k k
k
u u w RI t 2 21
2k k
k
w
spike arrival rate
constant input rates
no threshold
Next lecture:
11h25
![Page 20: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/20.jpg)
Biological Modeling
of Neural Networks:
Week 13 – Membrane potential
densities and Fokker-Planck
Wulfram Gerstner
EPFL, Lausanne, Switzerland
13.1 Review: Integrate-and-fire
- stochastic spike arrival
13.2 Density of membrane potential -
13.3 Flux - continuity equation
13.4. Fokker –Planck Equation
- -
13.5. Threshold and reset -
Week 13 – Membrane Potential Densities and Fokker-Planck Equation
![Page 21: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/21.jpg)
u
p(u)
u
Membrane potential density
22
2( , ) [ ( ) ( , )] ( , ) ( ) ( )resetp u t u p u t p u t A t u u
t u u
Fokker-Planck
drift diffusion ( ) k k
k
u u w RI 2 21
2k k
k
w
spike arrival rate
blackboard
Week 13-part 5: Threshold and reset (sink and source terms)
![Page 22: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/22.jpg)
u
p(u)
u
Membrane potential density
blackboard
Week 13-part 5: population firing rate A(t)
Population Firing rate A(t): flux at threshold
![Page 23: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/23.jpg)
I
0I)(tI
)()( tIRuuudt
deq
noiseo IItI )(
effective noise current
frequency f
0I
with noise
EPSC IPSC
Synaptic current pulses
)()()( ttIRuuudt
d meaneq
)()()(
','
''
,
fk
fki
fk
fkeeq ttqttqRuuu
dt
d
Week 13-part 5: population firing rate A(t) = single neuron rate
( )g I
![Page 24: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/24.jpg)
Week 13-part 5: membrane potential density
![Page 25: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/25.jpg)
Week 13-part 5: population activity, time-dependent
Nykamp+Tranchina,
2000
![Page 26: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/26.jpg)
Week 13-part 5: network states
frequency f
0I
with noise
( )g I
EPSC IPSC
( ) ( ) ( )mean
eq
du u u R I t t
dt
)()()(
','
''
,
fk
fki
fk
fkeeq ttqttqRuuu
dt
d
mean I(T) depends on state
Variance/noise depends on state
![Page 27: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/27.jpg)
Week 13-part 5: network states
Brunel 2000
![Page 28: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/28.jpg)
u
p(u)
Exercise 3: Diffusive noise + Threshold
u
Membrane potential density
sourcetupu
tupuu
tupt
),(
)],()([
),(
2
22
21
Fokker-Planck A= f
0I
with noise
- Calculate distribution p(u)
- Determine population firing rate A
Miniproject:
12h00
![Page 29: Biological Modeling - Continuity equation of Neural Networkslcn · Biological Modeling of Neural Networks: Week 13 – Membrane potential densities and Fokker-Planck Wulfram Gerstner](https://reader036.fdocuments.net/reader036/viewer/2022062604/5f83422109b4725f3e305b51/html5/thumbnails/29.jpg)
THE END
Sign up for:
- miniproject fraud detection interview