Feedforward and recurrent processing in cat primary visual cortex

This speaker is part of the Neuroscience Colloquium at the Department of Physics, Penn State University Dr. Miller is the Director of the Center for Theoretical Neuroscience at Columbia University. AbstractThe cat primary visual cortex (V1) receives visual inputs from the lateral geniculate nucleus (LGN).A great deal of recent evidence, and our modeling, suggest that fundamental response properties of neurons in the input-recipient layer of V1 are largely determined by the set of feedforward LGN inputs they receive, along with feedforward inhibition and cellular and synaptic properties. Yet these cells, like all V1 cells, receive large numbers of connections from other cortical cells._ What is the function of this strong recurrence? I will tell two stories that may shed some light on this. The first concerns surround suppression, and is joint work with the lab of David Ferster. A neuron's classical receptive field or "center" is the area in which an appropriate light stimulus can elicit spikes; stimuli in the "surround", outside the classical receptive field, cannot drive spikes but can modulate responses to center stimuli. I will show that measurements of surround suppression indicate that V1 operates as an inhibition-stabilized network (ISN), meaning that recurrent excitation alone is strong enough to be unstable by itself, but that the network is stabilized by feedback inhibition. Because an ISN is intrinsically stable despite strong recurrent excitation, it is compatible at least in principle with responses being determined by feedforward inputs, albeit with some cortical filtering. Second, I will show that any regime with strong recurrent excitation balanced by strong feedback inhibition has an unexpected dynamical correlate, namely strong transient amplification of certain patterns of activity without slowing of the dynamics associated with these patterns. This leads to structured amplification of variance, and in particular can explain certain observations of structure in the spontaneous activity of V1. Mathematically, this transient amplification is due to the strongly non-normal nature of the connection matrices that characterize the ISN architecture. _

Contact

  Dezhe Jin
  863-6673