Discussion Following the Talk
Q1: Why do you choose very specific kind of laminar circuits? What is the particular reason for ur circuits? Any predictions for us?
Answer: I would like you to refer to my paper with Razada. There we stimulate in layer 6 and observe what happens in layer 4.
The reason for Laminar architecture and how we came to it: We started working on grouping. What are the units of perception? To get at the grouping properties and analog coherence required certain ordering of mechanisms or else the data would collapse. Needed something which is robust and not parameter sensitive. We could see by analyzing the anatomy that it provided this required ordering of mechanisms in a compact way. We tried knocking down parts of it to see whether this breaks down and we found that the entire thing was necessary. What we basically have is a minimal mechanis for the required ordering property.
Qn:Do you insist that the lateral connections add superlinearly? Ans: Not superlinearly but faster.
Comment: The normalization property with just one inhibitory neron might not be sufficient for all. By variations of the simple circuit you can get a variety of effects from bipole to modulatory.
Qn : What is your prediction for an experiment where there is a stimulus in the center and the flanks are detected?
Qn: Does your model need spikes?
Answer: No you don't need spikes. Spikes are important in some situations (self syncrhonizing nets, order preserving limit cycles) but not required for other.
Qn: Why are spikes there?
Ans: So that we have non attenuated signal transmission over long distance. Also with spikes, we can compensate for axonal delay by increasing the diameter of axons.
Also I talked about a balance between excitation and inhibition. If these balances doesn't occur due to development then the system becomes to unstable. If the system is designed to 'stabilize' using spikes, then spikes become important. These systems show that they love to resynchronize these spikes for stability reasons.
Qn: Does your model do computer vision tasks that are not done yet?
For this the speaker answered with a list of computer vision tasks his algorithms have been employed on.
Qn: The neurophysiological detail is that there are spikes.
A: Why don't u put in all the channels?
Jeff (Comment): Analogy between computers and quantum mechanics. Although the transistors operate based on principles of quantum mechanics, an understanding of quantum mechanics is not required to understand the working of computers.
Speaker(Comment): My equations are mean values of stochastic differential equations.
Speaker(Comment): An example where spikes are not important: Phonemic Restoration. BB noise colored by
eel is on the
wheel wagon
peel orange
heel is on the shoes.
Pentti(Comment): Syncrhonous computing might not need spikes. Need not be the case for asynchronous operations.
Back to the
main index
for Inference and Prediction in Neocortical Circuits.