![Page 1: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/1.jpg)
Statistical analysis of spike trains in neuronalnetworks
Bruno Cessac
NeuroMathComp Team,INRIA Sophia Antipolis,France.
Mathematical Modeling and Statistical Analysis inNeuroscience, 02-07-14
Bruno Cessac Statistical analysis of spike trains in neuronal networks
![Page 2: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/2.jpg)
Bruno Cessac Statistical analysis of spike trains in neuronal networks
![Page 3: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/3.jpg)
Visual system
Bruno Cessac Statistical analysis of spike trains in neuronal networks
![Page 4: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/4.jpg)
Visual system
Bruno Cessac Statistical analysis of spike trains in neuronal networks
![Page 5: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/5.jpg)
Visual system
Bruno Cessac Statistical analysis of spike trains in neuronal networks
![Page 6: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/6.jpg)
Visual system
Bruno Cessac Statistical analysis of spike trains in neuronal networks
![Page 7: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/7.jpg)
Multi Electrodes Array
Figure : Multi-Electrodes Array.
![Page 8: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/8.jpg)
Multi Electrodes Array
![Page 9: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/9.jpg)
Encoding a visual scene
![Page 10: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/10.jpg)
Encoding a visual scene
![Page 11: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/11.jpg)
Encoding a visual scene
![Page 12: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/12.jpg)
Encoding a visual scene
![Page 13: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/13.jpg)
Encoding a visual scene
Do Ganglion cells act as independent encoders ?
Or do their dynamical (spatio-temporal) correlations play a role inencoding a visual scene (population coding) ?
![Page 14: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/14.jpg)
Encoding a visual scene
Do Ganglion cells act as independent encoders ?
Or do their dynamical (spatio-temporal) correlations play a role inencoding a visual scene (population coding) ?
![Page 15: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/15.jpg)
Let us measure (instantaneous pairwise) correlations
E. Schneidman, M.J. Berry, R. Segev, and W. Bialek. "Weak pairwise correlations imply strongly correlated
network states in a neural population". Nature, 440(7087):1007-1012, 2006.
![Page 16: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/16.jpg)
Let us measure (instantaneous pairwise) correlations
E. Schneidman, M.J. Berry, R. Segev, and W. Bialek. "Weak pairwise correlations imply strongly correlated
network states in a neural population". Nature, 440(7087):1007-1012, 2006.
![Page 17: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/17.jpg)
Let us measure (instantaneous pairwise) correlations
E. Schneidman, M.J. Berry, R. Segev, and W. Bialek. "Weak pairwise correlations imply strongly correlated
network states in a neural population". Nature, 440(7087):1007-1012, 2006.
![Page 18: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/18.jpg)
Constructing a statistical model handling measuredcorrelations
Assume stationarity.
Measure empirical correlations.
Select the probability distribution which maximizes theentropy and reproduces these correlations.
![Page 19: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/19.jpg)
Spike events
Figure : Spike state.
Spike state
!k(n) 2 f 0; 1 g
Spike pattern
!(n) = (!k(n) )Nk=1
Spike block
!nm = f!(m)!(m + 1) : : : !(n) g
Raster plot
!def= !T0
![Page 20: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/20.jpg)
Spike events
Figure : Spike pattern.
Spike state
!k(n) 2 f 0; 1 g
Spike pattern
!(n) = (!k(n) )Nk=1
Spike block
!nm = f!(m)!(m + 1) : : : !(n) g
Raster plot
!def= !T0
![Page 21: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/21.jpg)
Spike events
Figure : Spike pattern.
Spike state
!k(n) 2 f 0; 1 g
Spike pattern
!(n) = (!k(n) )Nk=1
�1001
�
Spike block
!nm = f!(m)!(m + 1) : : : !(n) g
Raster plot
!def= !T0
![Page 22: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/22.jpg)
Spike events
Figure : Spike block.
Spike state
!k(n) 2 f 0; 1 g
Spike pattern
!(n) = (!k(n) )Nk=1
Spike block
!nm = f!(m)!(m + 1) : : : !(n) g
Raster plot
!def= !T0
![Page 23: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/23.jpg)
Spike events
Figure : Spike block.
Spike state
!k(n) 2 f 0; 1 g
Spike pattern
!(n) = (!k(n) )Nk=1
Spike block
!nm = f!(m)!(m + 1) : : : !(n) g�1 1 0 1 00 1 0 1 00 0 0 1 00 1 0 1 1
�
Raster plot
!def= !T0
![Page 24: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/24.jpg)
Spike events
Figure : Raster plot/Spike train.
Spike state
!k(n) 2 f 0; 1 g
Spike pattern
!(n) = (!k(n) )Nk=1
Spike block
!nm = f!(m)!(m + 1) : : : !(n) g
Raster plot
!def= !T0
![Page 25: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/25.jpg)
Constructing a statistical model handling measuredcorrelations
Let �(T )! be the empirical measure:
�(T )! [ f ] =
1
T
TXt=1
f � �t(!)
e.g. �(T )! [!i ] =
1T
PTt=1 !i (t): �ring rate;
�(T )! [!i!j ] =
1T
PTt=1 !i (t)!j(t).
Find the (stationary) probability distribution � thatmaximizes statistical entropy under the constraints:
�(T )! [!i ] = �(!i );�
(T )! [!i!j ] = �(!i!j)
![Page 26: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/26.jpg)
Constructing a statistical model handling measuredcorrelations
There is a unique probability distribution which satis�es theseconditions.
This is the Gibbs distribution with potential:
H(!(0)) =NXi=1
hi!i (0) +NX
i ;j=1
Jij!i (0)!j(0)
Ising model
![Page 27: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/27.jpg)
End of the story ?
![Page 28: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/28.jpg)
End of the story ?
![Page 29: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/29.jpg)
End of the story ?
The Ising potential:
H(!(0)) =NXi=1
hi!i (0) +NX
i ;j=1
Jij!i (0)!j(0)
does not consider time correlations between neurons.It is therefore bad at predicting spatio-temporal patterns !
![Page 30: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/30.jpg)
Which correlations ?
Spikes correlations seem to play a role in spike coding.
Although this statement depends on several assumption that could bias statistics
Stationarity;
Binning;
Stimulus dependence ?
Modulo these remarks, Maximum entropy seems to be a relevantsetting to study the role of spatio-temporal spike correlations in
retina coding.
![Page 31: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/31.jpg)
Which correlations ?
Spikes correlations seem to play a role in spike coding.Although this statement depends on several assumption that could bias statistics
Stationarity;
Binning;
Stimulus dependence ?
Modulo these remarks, Maximum entropy seems to be a relevantsetting to study the role of spatio-temporal spike correlations in
retina coding.
![Page 32: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/32.jpg)
Which correlations ?
Spikes correlations seem to play a role in spike coding.Although this statement depends on several assumption that could bias statistics
Stationarity;
Binning;
Stimulus dependence ?
Modulo these remarks, Maximum entropy seems to be a relevantsetting to study the role of spatio-temporal spike correlations in
retina coding.
![Page 33: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/33.jpg)
OK. So let us consider spatio-temporal constraints.
![Page 34: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/34.jpg)
OK. So let us consider spatio-temporal constraints.
Easy !
Euh... In fact not so easy.
H(!D0 ) =NXi=1
hi!i (0) +NX
i ;j=1
J(0)ij !i (0)!j(0)
+NX
i ;j=1
J(1)ij !i (0)!j(1)
+NX
i ;j ;k=1
J(2)ijk !i (0)!j(1)!k(2)
+????
![Page 35: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/35.jpg)
OK. So let us consider spatio-temporal constraints.
Easy !
Euh... In fact not so easy.
H(!D0 ) =NXi=1
hi!i (0) +NX
i ;j=1
J(0)ij !i (0)!j(0)
+NX
i ;j=1
J(1)ij !i (0)!j(1)
+NX
i ;j ;k=1
J(2)ijk !i (0)!j(1)!k(2)
+????
![Page 36: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/36.jpg)
OK. So let us consider spatio-temporal constraints.
Easy !
Euh... In fact not so easy.
H(!D0 ) =NXi=1
hi!i (0) +NX
i ;j=1
J(0)ij !i (0)!j(0)
+NX
i ;j=1
J(1)ij !i (0)!j(1)
+NX
i ;j ;k=1
J(2)ijk !i (0)!j(1)!k(2)
+????
![Page 37: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/37.jpg)
OK. So let us consider spatio-temporal constraints.
Easy ! Euh... In fact not so easy.
H(!D0 ) =NXi=1
hi!i (0) +NX
i ;j=1
J(0)ij !i (0)!j(0)
+NX
i ;j=1
J(1)ij !i (0)!j(1)
+NX
i ;j ;k=1
J(2)ijk !i (0)!j(1)!k(2)
+????
![Page 38: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/38.jpg)
Two "small" problems.
Handling temporality and memory.
![Page 39: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/39.jpg)
Two "small" problems.
Handling temporality and memory.
![Page 40: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/40.jpg)
Two "small" problems.
Handling temporality and memory.
![Page 41: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/41.jpg)
Two "small" problems.
Handling temporality and memory.
Ising model considers successive times as independent
![Page 42: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/42.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 43: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/43.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 44: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/44.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 45: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/45.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 46: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/46.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 47: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/47.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 48: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/48.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 49: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/49.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 50: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/50.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 51: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/51.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
The probability of a spike pattern ....depends on the network history (transition probabilities).
![Page 52: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/52.jpg)
Two "small" problems.
Handling temporality and memory.
Probability of characteristic spatio-temporal patterns
Given a set of hypotheses on transition probabilities there exists amathematical framework to solve the problem.
![Page 53: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/53.jpg)
Handling memory.
Markov chains
Variable length Markov chains
Chains with complete connections
: : :
Gibbs distributions.
![Page 54: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/54.jpg)
Mathematical setting
Probability distribution on (bi-in�nite) rasters:� [!nm ] ;8m < n 2 Z
Conditional probabilities with memory depth D:Pn�!(n)
��!n�1n�D
�.
Generating arbitrary depth D blocks probabilities:
��!m+Dm
�= Pm+D
�!(m + D)
��!m+D�1m
���!m+D�1m
�� [!nm ] =
Qnl=m+D Pl
h!(l)
���!l�1l�D
i��!m+D�1m
�;
8m < n 2 ZChapman-Kolmogorov relation
![Page 55: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/55.jpg)
Mathematical setting
Probability distribution on (bi-in�nite) rasters:� [!nm ] ;8m < n 2 ZConditional probabilities with memory depth D:Pn�!(n)
��!n�1n�D
�.
Generating arbitrary depth D blocks probabilities:
��!m+Dm
�= Pm+D
�!(m + D)
��!m+D�1m
���!m+D�1m
�� [!nm ] =
Qnl=m+D Pl
h!(l)
���!l�1l�D
i��!m+D�1m
�;
8m < n 2 ZChapman-Kolmogorov relation
![Page 56: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/56.jpg)
Mathematical setting
Probability distribution on (bi-in�nite) rasters:� [!nm ] ;8m < n 2 ZConditional probabilities with memory depth D:Pn�!(n)
��!n�1n�D
�.
�1 1 0 1 j 00 1 0 1 j 00 0 0 1 j 00 1 0 1 j 1
�
Generating arbitrary depth D blocks probabilities:
��!m+Dm
�= Pm+D
�!(m + D)
��!m+D�1m
���!m+D�1m
�� [!nm ] =
Qnl=m+D Pl
h!(l)
���!l�1l�D
i��!m+D�1m
�;
8m < n 2 ZChapman-Kolmogorov relation
![Page 57: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/57.jpg)
Mathematical setting
Probability distribution on (bi-in�nite) rasters:� [!nm ] ;8m < n 2 ZConditional probabilities with memory depth D:Pn�!(n)
��!n�1n�D
�.
Generating arbitrary depth D blocks probabilities:
��!m+Dm
�= Pm+D
�!(m + D)
��!m+D�1m
���!m+D�1m
�� [!nm ] =
Qnl=m+D Pl
h!(l)
���!l�1l�D
i��!m+D�1m
�;
8m < n 2 ZChapman-Kolmogorov relation
![Page 58: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/58.jpg)
Mathematical setting
Probability distribution on (bi-in�nite) rasters:� [!nm ] ;8m < n 2 ZConditional probabilities with memory depth D:Pn�!(n)
��!n�1n�D
�.
Generating arbitrary depth D blocks probabilities:
��!m+Dm
�=
Pm+D
�!(m + D)
��!m+D�1m
���!m+D�1m
�� [!nm ] =
Qnl=m+D Pl
h!(l)
���!l�1l�D
i��!m+D�1m
�;
8m < n 2 ZChapman-Kolmogorov relation
![Page 59: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/59.jpg)
Mathematical setting
Probability distribution on (bi-in�nite) rasters:� [!nm ] ;8m < n 2 ZConditional probabilities with memory depth D:Pn�!(n)
��!n�1n�D
�.
Generating arbitrary depth D blocks probabilities:
��!m+Dm
�= Pm+D
�!(m + D)
��!m+D�1m
���!m+D�1m
�
� [!nm ] =Qn
l=m+D Pl
h!(l)
���!l�1l�D
i��!m+D�1m
�;
8m < n 2 ZChapman-Kolmogorov relation
![Page 60: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/60.jpg)
Mathematical setting
Probability distribution on (bi-in�nite) rasters:� [!nm ] ;8m < n 2 ZConditional probabilities with memory depth D:Pn�!(n)
��!n�1n�D
�.
Generating arbitrary depth D blocks probabilities:
��!m+Dm
�= Pm+D
�!(m + D)
��!m+D�1m
���!m+D�1m
�� [!nm ] =
Qnl=m+D Pl
h!(l)
���!l�1l�D
i��!m+D�1m
�;
8m < n 2 ZChapman-Kolmogorov relation
![Page 61: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/61.jpg)
Mathematical setting
� [!nm ] =nY
l=m+D
Pl
h!(l)
���!l�1l�D
i�h!m+D�1m
i; 8m < n 2 Z
�l
�!ll�D
�= logPl
h!(l)
���!l�1l�D
i
� [!nm ] = expnX
l=m+D
�l
�!ll�D
��h!m+D�1m
i
�h!nm j!m+D�1
m
i= exp
nXl=m+D
�l
�!ll�D
�
![Page 62: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/62.jpg)
Mathematical setting
� [!nm ] =nY
l=m+D
Pl
h!(l)
���!l�1l�D
i�h!m+D�1m
i; 8m < n 2 Z
�l
�!ll�D
�= logPl
h!(l)
���!l�1l�D
i
� [!nm ] = expnX
l=m+D
�l
�!ll�D
��h!m+D�1m
i
�h!nm j!m+D�1
m
i= exp
nXl=m+D
�l
�!ll�D
�
![Page 63: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/63.jpg)
Mathematical setting
� [!nm ] =nY
l=m+D
Pl
h!(l)
���!l�1l�D
i�h!m+D�1m
i; 8m < n 2 Z
�l
�!ll�D
�= logPl
h!(l)
���!l�1l�D
i
� [!nm ] = expnX
l=m+D
�l
�!ll�D
��h!m+D�1m
i
�h!nm j!m+D�1
m
i= exp
nXl=m+D
�l
�!ll�D
�
![Page 64: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/64.jpg)
Mathematical setting
� [!nm ] =nY
l=m+D
Pl
h!(l)
���!l�1l�D
i�h!m+D�1m
i; 8m < n 2 Z
�l
�!ll�D
�= logPl
h!(l)
���!l�1l�D
i
� [!nm ] = expnX
l=m+D
�l
�!ll�D
��h!m+D�1m
i
�h!nm j!m+D�1
m
i= exp
nXl=m+D
�l
�!ll�D
�
![Page 65: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/65.jpg)
Gibbs distribution
8� � Zd ; �(f S g j @�) = 1
Z�;@�e��H
�;@�( fS g )
f (�) = � 1
�lim�"1
1
j�j logZ�;@�
(free energy density)
![Page 66: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/66.jpg)
Gibbs distribution
8� � Zd ; �(f S g j @�) = 1
Z�;@�e��H
�;@�( fS g )
f (�) = � 1
�lim�"1
1
j�j logZ�;@�
(free energy density)
![Page 67: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/67.jpg)
Gibbs distribution
8� � Zd ; �(f S g j @�) = 1
Z�;@�e��H
�;@�( fS g )
f (�) = � 1
�lim�"1
1
j�j logZ�;@�
(free energy density)
![Page 68: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/68.jpg)
Gibbs distribution
![Page 69: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/69.jpg)
Gibbs distribution
8m; n; �h!nm j!m+D�1
m
i= exp
nXl=m+D
�l
�!ll�D
�
(normalized potential)
![Page 70: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/70.jpg)
Gibbs distribution
8m < n; A <� [!nm ]
expPn
l=m+DH�!ll�D
�exp�(n �m)P(H)
< B
(non normalized potential)
![Page 71: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/71.jpg)
Gibbs distribution
P(H) is called "topological pressure" and is formalyequivalent to free energy density.
Does not require time-translation invariance (stationarity).
In the stationary case (+ assumptions) a Gibbs state is alsoan equilibrium state.
sup�2Minv
h(�) + �(H) = h(�) + �(H) = P(H)
.
![Page 72: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/72.jpg)
Gibbs distribution
This formalism allows to handle the spatio-temporal case
H(!D0 ) =NXi=1
hi!i (0) +NX
i ;j=1
J(0)ij !i (0)!j(0)
+NX
i ;j=1
J(1)ij !i (0)!j(1)
+NX
i ;j ;k=1
J(2)ijk !i (0)!j(1)!k(2) + : : :
even numerically.J.C. Vasquez, A. Palacios, O. Marre, M.J. Berry II, B. Cessac, J. Physiol. Paris, , Vol 106, Issues 34, (2012).
H. Nasser, O. Marre, and B. Cessac, J. Stat. Mech. (2013) P03006.
H. Nasser, B. Cessac, Entropy (2014), 16(4), 2244-2277.
![Page 73: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/73.jpg)
Gibbs distribution
This formalism allows to handle the spatio-temporal case
H(!D0 ) =NXi=1
hi!i (0) +NX
i ;j=1
J(0)ij !i (0)!j(0)
+NX
i ;j=1
J(1)ij !i (0)!j(1)
+NX
i ;j ;k=1
J(2)ijk !i (0)!j(1)!k(2) + ?????
even numerically.J.C. Vasquez, A. Palacios, O. Marre, M.J. Berry II, B. Cessac, J. Physiol. Paris, , Vol 106, Issues 34, (2012).
H. Nasser, O. Marre, and B. Cessac, J. Stat. Mech. (2013) P03006.
H. Nasser, B. Cessac, Entropy (2014), 16(4), 2244-2277.
![Page 74: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/74.jpg)
Two small problems.
Exponential number of possible terms.
Contrarily to what happens usually in physics, we do not knowwhat should be
the right potential.
![Page 75: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/75.jpg)
Two small problems.
Exponential number of possible terms.
Contrarily to what happens usually in physics, we do not knowwhat should be
the right potential.
![Page 76: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/76.jpg)
Can we have a reasonable idea of what could be the spike statisticsby studying a neural network model ?
![Page 77: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/77.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
![Page 78: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/78.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
R.Cofr�e,B. Cessac: "Dynamics and spike trains statistics in conductance-based Integrate-and-Fire neural networks
with chemical and electric synapses", Chaos, Solitons and Fractals, 2013.
![Page 79: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/79.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
Sub-threshold dynamics:
CkdVk
dt= �gL;k(Vk � EL)
�Xj
gkj(t; !)(Vk � Ej)
�Xj
�gkj (Vk � Vj)
+i(ext)k (t) + �B�k(t)
![Page 80: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/80.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
Sub-threshold dynamics:
CkdVk
dt= �gL;k(Vk � EL)
�Xj
gkj(t; !)(Vk � Ej)
�Xj
�gkj (Vk � Vj)
+i(ext)k (t) + �B�k(t)
![Page 81: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/81.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
Sub-threshold dynamics:
CkdVk
dt= �gL;k(Vk � EL)
�Xj
gkj(t; !)(Vk � Ej)
�Xj
�gkj (Vk � Vj)
+i(ext)k (t) + �B�k(t)
![Page 82: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/82.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
Sub-threshold dynamics:
CkdVk
dt= �gL;k(Vk � EL)
�Xj
gkj(t; !)(Vk � Ej)
�Xj
�gkj (Vk � Vj)
+i(ext)k (t) + �B�k(t)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0.5 1 1.5 2 2.5 3 3.5 4
PSP
t
t=1t=1.2t=1.6
t=3g(x)
![Page 83: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/83.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
Sub-threshold dynamics:
CkdVk
dt= �gL;k(Vk � EL)
�Xj
gkj(t; !)(Vk � Ej)
�Xj
�gkj (Vk � Vj)
+i(ext)k (t) + �B�k(t)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0.5 1 1.5 2 2.5 3 3.5 4
PSP
t
g(x)
![Page 84: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/84.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
Sub-threshold dynamics:
CkdVk
dt= �gL;k(Vk � EL)
�Xj
gkj(t; !)(Vk � Ej)
�Xj
�gkj (Vk � Vj)
+i(ext)k (t) + �B�k(t)
![Page 85: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/85.jpg)
An Integrate and Fire neural network model with chemicaland electric synapses
Sub-threshold dynamics:
CkdVk
dt= �gL;k(Vk � EL)
�Xj
gkj(t; !)(Vk � Ej)
�Xj
�gkj (Vk � Vj)
+i(ext)k (t) + �B�k(t)
![Page 86: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/86.jpg)
Sub-threshold regime
CdV
dt+�G (t; !)� G
�V = I (t; !);
Gkl(t; !) =
24 gL;k +
NXj=1
gkj(t; !)
35 �kl
def= gk(t; !)�kl :
I (t; !) = I (cs)(t; !) + I (ext)(t) + I (B)(t)
I(cs)k (t; !) =
Xj
Wkj�kj(t; !); Wkjdef= GkjEj :
![Page 87: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/87.jpg)
Sub-threshold regime
8<:
dV = (�(t; !)V + f (t; !))dt + �BcINdW (t);
V (t0) = v ;
�(t; !) = C�1�G � G (t; !)
�
f (t; !) = C�1I (cs)(t; !) + C�1I (ext)(t)
![Page 88: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/88.jpg)
Homogeneous Cauchy problem
�dV (t;!)
dt= �(t; !)V (t; !);
V (t0) = v ;
Theorem
�(t; !) square matrix with bounded elements.
M0(t0; t; !) = IN
Mk(t0; t; !) = IN +
Z t
t0
�(s; !)Mk�1(s; t)ds; t � t1;
converges uniformly in [t0; t1].Brockett, R. W., "Finite Dimensional Linear Systems",John Wiley and Sons, 1970.
Flow
�(t0; t; !)def= lim
k!1Mk(t0; t; !)
![Page 89: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/89.jpg)
Homogeneous Cauchy problem
�dV (t;!)
dt= �(t; !)V (t; !);
V (t0) = v ;
Theorem
�(t; !) square matrix with bounded elements.
M0(t0; t; !) = IN
Mk(t0; t; !) = IN +
Z t
t0
�(s; !)Mk�1(s; t)ds; t � t1;
converges uniformly in [t0; t1].Brockett, R. W., "Finite Dimensional Linear Systems",John Wiley and Sons, 1970.
Flow
�(t0; t; !)def= lim
k!1Mk(t0; t; !)
![Page 90: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/90.jpg)
Homogeneous Cauchy problem
�dV (t;!)
dt= �(t; !)V (t; !);
V (t0) = v ;
Theorem
�(t; !) square matrix with bounded elements.
M0(t0; t; !) = IN
Mk(t0; t; !) = IN +
Z t
t0
�(s; !)Mk�1(s; t)ds; t � t1;
converges uniformly in [t0; t1].Brockett, R. W., "Finite Dimensional Linear Systems",John Wiley and Sons, 1970.
Flow
�(t0; t; !)def= lim
k!1Mk(t0; t; !)
![Page 91: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/91.jpg)
Homogeneous Cauchy problem
If �(t; !) and �(s; !) commute
�(t0; t; !) =1Xk=0
1
k!(
Z t
t0
�(s; !)ds)k = eR tt0�(s;!)ds
![Page 92: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/92.jpg)
Homogeneous Cauchy problem
If �(t; !) and �(s; !) commute
�(t0; t; !) =1Xk=0
1
k!(
Z t
t0
�(s; !)ds)k = eR tt0�(s;!)ds
This holds only in two cases :
G = 0;
�(t0; t; !) = e� 1
c
R tt0G(s;!)ds
B. Cessac, J. Math. Neuroscience, 2011.
![Page 93: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/93.jpg)
Homogeneous Cauchy problem
If �(t; !) and �(s; !) commute
�(t0; t; !) =1Xk=0
1
k!(
Z t
t0
�(s; !)ds)k = eR tt0�(s;!)ds
This holds only in two cases :
G = 0;
�(t0; t; !) = e� 1
c
R tt0G(s;!)ds
B. Cessac, J. Math. Neuroscience, 2011.
G (t; !) = �(t; !)IN
![Page 94: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/94.jpg)
Homogeneous Cauchy problem
In general:
�(t0; t; !) = IN++1Xn=1
XX1 = ( B; A(s1; !) )X2 = ( B; A(s2; !) )
: : :
Xn = ( B; A(sn; !) )
Z t
t0
� � �Z sn�1
t0
nYk=1
Xk ds1 � � � dsn :
B = C�1G ; A(t; !) = �C�1G (t; !)
![Page 95: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/95.jpg)
Exponentially bounded ow
De�nition: An exponentially bounded ow is a two parameter(t0; t) family f�(t0; t; !)gt�t0 of ows such that, 8! 2 :
1 �(t0; t0; !) = IN and �(t0; t; !)�(t; s; !) = �(t0; s; !)whenever t0 � t � s;
2 For each v 2 RN and ! 2 , (t0; t)! �(t0; t; !)v iscontinuous for t0 � t;
3 There is M > 0 and m > 0 such that :
jj�(s; t; !)jj � Me�m(t�s); s � t: (1)
![Page 96: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/96.jpg)
Exponentially bounded ow
Proposition
Let �1 be the largest eigenvalue of �G. If:
�1 < gL;
then the ow � in our model has the exponentially bounded ow
property.
Remark The typical electrical conductance values are of order 1nano-Siemens, while the leak conductance of retinal ganglion cellsis of order 50 micro-Siemens. Therefore, this condition is
compatible with the biophysical values of conductances in the
retina.
![Page 97: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/97.jpg)
Exponentially bounded ow
Proposition
Let �1 be the largest eigenvalue of �G. If:
�1 < gL;
then the ow � in our model has the exponentially bounded ow
property.
Remark The typical electrical conductance values are of order 1nano-Siemens, while the leak conductance of retinal ganglion cellsis of order 50 micro-Siemens. Therefore, this condition is
compatible with the biophysical values of conductances in the
retina.
![Page 98: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/98.jpg)
Exponentially bounded ow
Theorem
If �(t0; t; !) is an exponentially bounded ow , there is a unique
strong solution for t � t0 given by:
V (t0; t; !) = �(t0; t; !)v+
Z t
t0
�(s; t; !)f (s; !)ds+�B
c
Z t
t0
�(s; t; !)dW (s):
R. Wooster, "Evolution systems of measures for non-autonomous stochastic di�erential equations with Levy noise",
Communications on Stochastic Analysis, vol 5, 353-370, 2011
![Page 99: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/99.jpg)
Membrane potential decomposition
V (t; !) = V (d)(t; !) + V (noise)(t; !);
V (d)(t; !) = V (cs)(t; !) + V (ext)(t; !);
V (cs)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (cs)(s; !)ds;
V (ext)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (ext)(s; !)ds;
V (noise)(t; !) =�B
c
Z t
�k (t;!)�(s; t; !) dW (s):
![Page 100: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/100.jpg)
Membrane potential decomposition
V (t; !) = V (d)(t; !) + V (noise)(t; !);
V (d)(t; !) = V (cs)(t; !) + V (ext)(t; !);
V (cs)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (cs)(s; !)ds;
V (ext)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (ext)(s; !)ds;
V (noise)(t; !) =�B
c
Z t
�k (t;!)�(s; t; !) dW (s):
![Page 101: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/101.jpg)
Membrane potential decomposition
V (t; !) = V (d)(t; !) + V (noise)(t; !);
V (d)(t; !) = V (cs)(t; !) + V (ext)(t; !);
V (cs)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (cs)(s; !)ds;
V (ext)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (ext)(s; !)ds;
V (noise)(t; !) =�B
c
Z t
�k (t;!)�(s; t; !) dW (s):
![Page 102: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/102.jpg)
Membrane potential decomposition
V (t; !) = V (d)(t; !) + V (noise)(t; !);
V (d)(t; !) = V (cs)(t; !) + V (ext)(t; !);
V (cs)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (cs)(s; !)ds;
V (ext)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (ext)(s; !)ds;
V (noise)(t; !) =�B
c
Z t
�k (t;!)�(s; t; !) dW (s):
![Page 103: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/103.jpg)
Membrane potential decomposition
V (t; !) = V (d)(t; !) + V (noise)(t; !);
V (d)(t; !) = V (cs)(t; !) + V (ext)(t; !);
V (cs)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (cs)(s; !)ds;
V (ext)(t; !) =1
c
Z t
�1�(s; t; !)�(s; !) I (ext)(s; !)ds;
V (noise)(t; !) =�B
c
Z t
�k (t;!)�(s; t; !) dW (s):
![Page 104: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/104.jpg)
Transition probabilities
Pb: to determine P�!(n)
��!n�1�1
�
Fix !, n and t < n. Set:
b�k(t; !) = � � V(d)k (t; !); (1)
Neuron k emits a spike at integer time n (!k(n) = 1) if:
9t 2 [n � 1; n]; V(noise)k (t; !) = b�k(t; !):
"First passage" problem, in N dimension, with a time
dependent boundary b�k(t; !). (general form unknown).
![Page 105: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/105.jpg)
Transition probabilities
Pb: to determine P�!(n)
��!n�1�1
�Fix !, n and t < n. Set:
b�k(t; !) = � � V(d)k (t; !); (1)
Neuron k emits a spike at integer time n (!k(n) = 1) if:
9t 2 [n � 1; n]; V(noise)k (t; !) = b�k(t; !):
"First passage" problem, in N dimension, with a time
dependent boundary b�k(t; !). (general form unknown).
![Page 106: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/106.jpg)
Transition probabilities
Pb: to determine P�!(n)
��!n�1�1
�Fix !, n and t < n. Set:
b�k(t; !) = � � V(d)k (t; !); (1)
Neuron k emits a spike at integer time n (!k(n) = 1) if:
9t 2 [n � 1; n]; V(noise)k (t; !) = b�k(t; !):
"First passage" problem, in N dimension, with a time
dependent boundary b�k(t; !). (general form unknown).
![Page 107: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/107.jpg)
Transition probabilities
Pb: to determine P�!(n)
��!n�1�1
�Fix !, n and t < n. Set:
b�k(t; !) = � � V(d)k (t; !); (1)
Neuron k emits a spike at integer time n (!k(n) = 1) if:
9t 2 [n � 1; n]; V(noise)k (t; !) = b�k(t; !):
"First passage" problem, in N dimension, with a time
dependent boundary b�k(t; !). (general form unknown).
![Page 108: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/108.jpg)
Conditional probability
Without electric synapses the probability of !(n) conditionally to!n�1�1 can be approximated by:
P�!(n)
��!n�1�1
�=
NYk=1
P�!k(n)
��!n�1�1
�;
with P�!k(n)
��!n�1�1
�=
!k(n)� (Xk(n � 1; !)) + (1� !k(n)) (1� � (Xk(n � 1; !))) ;
where
Xk(n � 1; !) =� � V
(det)k (n � 1; !)
�k(n � 1; !);
and
�(x) =1p2�
Z +1
x
e�u2
2 du:
![Page 109: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/109.jpg)
Conditional probability
�(!) = logP�!(n)
��!n�1�1
�de�nes a (in�nite range)
normalized potential de�ning a unique Gibbs distribution.
It depends explicitly on networks parameters and external
stimulus.
Its de�nition holds for a time-dependent stimulus (nonstationary).
It is similar to the so-called Generalized Linear Model used forretina analysis, although with a more complex structure.
The general form (with electric synapses) is yet unknown.
![Page 110: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/110.jpg)
Conditional probability
�(!) = logP�!(n)
��!n�1�1
�de�nes a (in�nite range)
normalized potential de�ning a unique Gibbs distribution.
It depends explicitly on networks parameters and external
stimulus.
Its de�nition holds for a time-dependent stimulus (nonstationary).
It is similar to the so-called Generalized Linear Model used forretina analysis, although with a more complex structure.
The general form (with electric synapses) is yet unknown.
![Page 111: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/111.jpg)
Conditional probability
�(!) = logP�!(n)
��!n�1�1
�de�nes a (in�nite range)
normalized potential de�ning a unique Gibbs distribution.
It depends explicitly on networks parameters and external
stimulus.
Its de�nition holds for a time-dependent stimulus (nonstationary).
It is similar to the so-called Generalized Linear Model used forretina analysis, although with a more complex structure.
The general form (with electric synapses) is yet unknown.
![Page 112: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/112.jpg)
Conditional probability
�(!) = logP�!(n)
��!n�1�1
�de�nes a (in�nite range)
normalized potential de�ning a unique Gibbs distribution.
It depends explicitly on networks parameters and external
stimulus.
Its de�nition holds for a time-dependent stimulus (nonstationary).
It is similar to the so-called Generalized Linear Model used forretina analysis, although with a more complex structure.
The general form (with electric synapses) is yet unknown.
![Page 113: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/113.jpg)
Conditional probability
�(!) = logP�!(n)
��!n�1�1
�de�nes a (in�nite range)
normalized potential de�ning a unique Gibbs distribution.
It depends explicitly on networks parameters and external
stimulus.
Its de�nition holds for a time-dependent stimulus (nonstationary).
It is similar to the so-called Generalized Linear Model used forretina analysis, although with a more complex structure.
The general form (with electric synapses) is yet unknown.
![Page 114: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/114.jpg)
Back to our second "small" problem
Is there a Maximum Entropy potential corresponding to � (inthe stationary case) ?
![Page 115: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/115.jpg)
Back to our second "small" problem
Is there a Maximum Entropy potential corresponding to � (inthe stationary case) ?
![Page 116: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/116.jpg)
Back to our second "small" problem
One can make a Taylor expansion of �(!).
![Page 117: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/117.jpg)
Back to our second "small" problem
Using !i (n)k = !i (n); k � 1 one ends up with a potential of the
form:
�(!) =NXi=1
hi!i (0) +NX
i ;j=1
J(0)ij !i (0)!j(0) + : : :
![Page 118: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/118.jpg)
Back to our second "small" problem
The expansion is in�nite although one can approximate the in�niterange potential � by a �nite range approximation (�nite memory),
giving rise to a �nite expansion.
![Page 119: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/119.jpg)
Back to our second "small" problem
The coe�cients of the expansion are non linear functions of thenetwork parameters and stimulus.
They are therefore somewhat redundant.
![Page 120: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/120.jpg)
Back to our second "small" problem
Rodrigo Cofr�e, Bruno Cessac, "Exact computation of the maximum-entropy potential of spiking neural-network
models",Phys. Rev. E 89, 052117.
Given a set of stationary transition probabilities P�!(D)
��!D�1
0
�> 0
there is a unique (up to a constant) Maximum Entropy potential, written
as a linear combination of spike interactions terms with a minimal
number of terms (normal form). This potential can be explicitly (and
algorithmically) computed.
Hints: Using variable change one can eliminate terms in thepotential ("normal" form).
The construction is based on equivalence between Gibbs potentials(cohomology) and periodic orbits expansion.
![Page 121: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/121.jpg)
Back to our second "small" problem
However, there is still a number of terms growing exponentially
with the number of neurons and the memory depth.
These terms are generically non zero.
![Page 122: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/122.jpg)
Back to the retina
Neuromimetic models have typically O(N2) parameters whereN is the number of neurons.
The equivalent MaxEnt potential has generically a number ofparameters growing exponentially with N, non linear andredundant functions of the network parameters (synapticweights, stimulus).
)Intractable determination of parameters;
Stimulus dependent parameters;
Over�tting.
BUT Real neural networks are not generic
![Page 123: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/123.jpg)
Back to the retina
Neuromimetic models have typically O(N2) parameters whereN is the number of neurons.
The equivalent MaxEnt potential has generically a number ofparameters growing exponentially with N, non linear andredundant functions of the network parameters (synapticweights, stimulus).
)Intractable determination of parameters;
Stimulus dependent parameters;
Over�tting.
BUT Real neural networks are not generic
![Page 124: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/124.jpg)
Back to the retina
Neuromimetic models have typically O(N2) parameters whereN is the number of neurons.
The equivalent MaxEnt potential has generically a number ofparameters growing exponentially with N, non linear andredundant functions of the network parameters (synapticweights, stimulus).
)Intractable determination of parameters;
Stimulus dependent parameters;
Over�tting.
BUT Real neural networks are not generic
![Page 125: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/125.jpg)
Back to the retina
Neuromimetic models have typically O(N2) parameters whereN is the number of neurons.
The equivalent MaxEnt potential has generically a number ofparameters growing exponentially with N, non linear andredundant functions of the network parameters (synapticweights, stimulus).
)Intractable determination of parameters;
Stimulus dependent parameters;
Over�tting.
BUT
Real neural networks are not generic
![Page 126: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/126.jpg)
Back to the retina
Neuromimetic models have typically O(N2) parameters whereN is the number of neurons.
The equivalent MaxEnt potential has generically a number ofparameters growing exponentially with N, non linear andredundant functions of the network parameters (synapticweights, stimulus).
)Intractable determination of parameters;
Stimulus dependent parameters;
Over�tting.
BUT Real neural networks are not generic
![Page 127: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/127.jpg)
Back to the retina
MaxEnt approach might be useful if there is some hidden law ofnature/ symmetry which cancels most terms in the expansion.
![Page 128: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/128.jpg)
Acknowledgment
Neuromathcomp team
Rodrigo Cofr�e (pHd, September 2014)
Dora Karvouniari (M2)
Pierre Kornprobst (CR1 INRIA)
Slim Kraria (IR)
Gaia Lombardi (M2! Paris)
Hassan Nasser (pHd! Startup)
Daniela Pamplona (PostDoc)
Geo�rey Portelli (Post Doc)
Vivien Robinet (Post Doc! MCF Kourou)
Horacio Rostro (pHd! Docent Mexico)
Wahiba Taouali (Post Doc! Post Doc INTMarseille)
Juan-Carlos Vasquez (pHd! Post Doc Bogota)
Princeton University
Michael J. Berry II
ANR KEOPS
Maria-Jos�e Escobar (CN Valparaiso)
Adrian Palacios (CN Valparaiso)
Cesar Ravelo (CN Valparaiso)
Thierry Vi�eville (INRIA Mnemosyne)
Renvision FP7 project
Luca Bernondini (IIT Genova)
Matthias Hennig (Edinburgh)
Alessandro Maccionne (IIT Genova)
Evelyne Sernagor (Newcastle)
Institut de la Vision
Olivier Marre
Serge Picaud
Bruno Cessac Statistical analysis of spike trains in neuronal networks
![Page 129: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/129.jpg)
Can we hear the shape of a Maximum entropy potential
Two distinct potentials H(1);H(2) of range R = D + 1 correspondto the same Gibbs distribution (are \equivalent"), if and only ifthere exists a range D function f such that (Chazottes-Keller(2009)):
H(2)�!D0
�= H(1)
�!D0
�� f
�!D�10
�+ f
�!D1
�+�; (2)
where � = P(H(2))� P(H(1)).
![Page 130: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/130.jpg)
Can we hear the shape of a Maximum entropy potential
Summing over periodic orbits we get rid of the function f
RXn=1
�(!�nl1) =RXn=1
H�(!�nl1)� RP(H�); (3)
We eliminate equivalent constraints.
![Page 131: Statistical analysis of spike trains in neuronal networks · 2017. 1. 28. · Statistical analysis of spike trains in neuronal networks Bruno Cessac NeuroMathComp eam,INRIAT Sophia](https://reader034.vdocument.in/reader034/viewer/2022052007/601b61fadd9a51190a2b996f/html5/thumbnails/131.jpg)
Can we hear the shape of a Maximum entropy potential
Conclusion
Given a set of transition probabilities Ph!(D)
���!D�10
i> 0 there
is a unique, up to a constant, MaxEnt potential, written as a linearcombination of constraints (average of spike events) with aminimal number of terms. This potential can be explicitly (andalgorithmically) computed.