Skip to content. | Skip to navigation

Sections
Personal tools
You are here: Home → Forum → Using data sets → pvc8 - Response equation in paper and relation with code
 

pvc8 - Response equation in paper and relation with code

Up to Using data sets

pvc8 - Response equation in paper and relation with code

Posted by Paloma Casteleiro Costa at April 15. 2017

Hello, 

We are Georgia Tech students analyzing and wanting to fully understand this paper and the codes. 

We were wondering if someone could give us some hints about the response equations and how the parameters are tuned and chosen. 

- Where in the code are these parameters (beta, alpha, omega, gamma and n) chosen and optimized?

- Could someone give us some clarification on the difference between Erf and Ek in the equation. It isn't very clear, we have an idea but are trying to get the accurate difference.

 

Thank you so much!

 

Re: pvc8 - Response equation in paper and relation with code

Posted by Ruben Coen-Cagli at April 26. 2017

Hi, 

thanks for your interest in our data and code. 

In the 2015 Nature Neuro paper, we don't use the actual MGSM outputs to fit neural data, we only use the probability that the image patch is homogeneous (eq 2 in the 2015 paper). To fit data, we use the descriptive model of eq 1 in the 2015 paper. To compute R according to eq 1, you can compute Erf, Ek, Es from the outputs of the filters applied to the experimental images, as described in Methods:
"The drive to the RF, Erf, is defined as follows. The RF of each cell was represented by a pair of linear filters (quadrature pair, of even and odd phase) [NOTE: both filters of the pair have the same preferred orientation and spatial frequency]. The output of each filter was defined by the dot product between the image and the filter. The drive was then defined as the square root of the sum of squares of the filters’ outputs.
[...] The normalization signal in the RF, Ek, was computed from the output of four quadrature pairs of filters. These filters had the same size, position, and spatial frequency tuning as those of ERF , but each pair had a different preferred orientation (0, 45, 90 and 135 degrees), yielding a normalization signal that was untuned for orientation." 
 
Then  alpha, beta, gamma, & sigma (and the exponent 'n') are free parameters that you use to fit the neural data. These parameters do not appear in the MGSM model or code.
 
I hope this helps. 
Powered by Ploneboard
Document Actions