Talk Summary:
The speaker presented a spike-timing based communication mechanism for inter-neuron communication. He suggested reasons why a rate-coding based approach will not work. The proposed model includes a clock of period 20 msec. A signal P to be transmitted is encoded as the multiplication of a probability of firing sqrt(P) and a phase offset of sqrt(P) with respect to the clock. This model can reproduce observed PSTHs and receptive field characteristics. The 'Distributed Synchrony Model' was motivated by three criteria (1) Fast, reliable inter-cortical communication, (2) the need for a cell to multiplex and (3) the need to reproduce observed cell responses.
Discussion and Questions
Q: Is the cortex trying to reconstruct the image or trying to extract some features from the image?
A: In this case yes..trying to reconstruct.
Q: (Regarding the PSTHs). Your model will reproduce the PSTHs only when the stimulus is presented in synchronization with the internal clock.
A: That is true. But we assume that the cortex picks up the signal from the retina in a syncrhonized fashion.
Q: (Regarding) Coding cost of different strategies. Does the cost come from the learning part or the prediction part?
A: We don't make that distinction.
Q: Does your feedback rule have any relation with EM?
A: The feedback rule is motivated by the EM rule.
Q: (Regarding the synaptic weights used for receptive field simulations) Are the synaptic weights pretty similar or are they different.
A: Depending on how we pick the weights we can get different receptive fields
(Steve): You have a shut up model. But in my talk I gave an example where top-down feedback is excitatory. Could you comment on all these experiments that say that feedback is not shut up.
A: In this model feedback is indeed shutup.
Comment(Alessandra): Anatomical studies show that feedback is excitatory in center and inhibitory in surround.
Comment(Steve): I thinkt that end stopping is caused by feedback from V2 is false.
Q: It seems that phase has done two things with the clock. Multiplexing and Coding.
A: That is right. The width of the 20 ms is related to the number of 'threads' that you have.
Q(Bill): At the beginning of the clock..how does the neuron forget everything?
A: We assume that there is some magical swap space available for storing all the information and then swapping it back in.
Q(Bill): Also how does the cell generate the random number?
A: There are some known mechanisms.
Q: Are spikes are essential?
A: I think spikes are an innovation by biology to send signals reliably over long distance to
Q: If we have a package of spikes ....what is the necessity of clock?
A: So that we can do feedback. If there is no clock, there is no way to align the feedback signal.
Q: How neurons syncrhonize critically depend on the neuron model. For eg the Aertson model
A: He has an integrate and fire in combination with a random number generator.
Back to the
main index
for Inference and Prediction in Neocortical Circuits.