1 of 25

The neural code�Point processes

Kenneth D Harris, UCL

2 of 25

Point processes

  • A point process defines a probability distribution over the space of possible spike trains

Sample space =

all possible spike trains

 

3 of 25

The Poisson process

  •  

4 of 25

Spike counts in the Poisson process

  •  

5 of 25

Spike count histograms: Allen cell 180 (visual cortex )

6 of 25

Spike count histograms: Allen cell 181 (visual cortex )

7 of 25

Spike count histograms: Allen cell 0 (dentage gyrus )

  • Doesn’t look like a Poisson distribution!

8 of 25

Autocorrelogram and ISI histogram

  •  

9 of 25

ISI and autocorrelogram examples (IBL superior colliculus cells)

10 of 25

Inhomogeneous Poisson process

  •  

Local field potential

Time

Time

Intensity

11 of 25

Model selection by spike train prediction

Training set

 

Test set

Evaluate prediction

12 of 25

Measuring prediction quality: likelihood

  •  

 

13 of 25

Information theory interpretation

  •  

14 of 25

Measuring prediction quality: squared error

  •  

slope=Q

15 of 25

Interpretation of Q

  •  

 

 

 

 

 

16 of 25

Example: hippocampal place cells

17 of 25

Summary across cells

18 of 25

Summary across cells

19 of 25

Summary across cells

20 of 25

Summary across cells

21 of 25

Summary across cells

22 of 25

Predicting spike train from events

  •  

23 of 25

Fitting models with a Toeplitz matrix

 

*

=

 

 

24 of 25

Fitting kernels

  •  

25 of 25

Leave kernels out to see if you needed them