Feed-forward model


Trying to train models that have an initial bias with Arani's stims

Issue: The initial feed-forward bias stimulus that we are using in the feed-forward modeling paper Van Hooser et al. (2014) responds differently to Arani's stimuli than cells in the real animal

This suggests that IF a feed-forward structure is in the brain, it likely does not start out "flat", with all locations equal.

This means that our neurons behave more like cells that have an "X" shaped kernel

Wes - How to do it:

use the function


where kernel is defined as follows:

A = eye(8); % a diagonal of 1's

kernel = 2e-9*(A+A(end:-1:1,:)); % an 'X' configuration

This will train a feedforward network with an X kernel configuration on stimulus number 7.

You'll want to play around with a few parameters:

1) the starting magnitude of the kernel -- if it is too big you'll just get a ton of spikes that amplifies

2) the maximum synaptic weight; if this is too big then the network will spike to everything

Check out the documentation to train_Bsequences to see the input arguments. I'd recommend plotting the simulations at first until you've found a nice parameter range that just gives a few spikes to sequence 7 and sequence 1 (maybe differs by a factor of 2-3). Then you can start training with, say, 100 trials.