next up previous
Next: SmirnakisWarland, Berry, Up: No Title Previous: Shadlen and Newsome:

Softky: Thalamus as a predictive filter


Thalamocortical ``relay'' cells--which transmit sensory signals to primary cortical areas--also receive massive feedback from cortex, both direct excitation and indirect inhibition. What function might such push-pull feedback serve?

Suppose that the cumulative feedback to a given relay cell provides a particular context signal: it tends to predict the cell's imminent sensory input. Then that feedback could perform either of two opposite functions. Excitatory predictive feedback would bolster the Relay cell's spiking when it is already expected to fire, thereby enhancing its reliable response to noisy sensory input. Inhibitory predictive feedback, on the other hand, would suppress predictable firing, only letting the cell fire when expectation and reality disagreed (i.e. when the input carried novel information not already available from context). Because a cell in this latter mode would signal the difference between expected and actual sensory input, its firing would need to convey distinct positive and negative signals along the same axon: (positive) unpredicted input through single spikes, and (negative) unfulfilled predictions through post-inhibitory-rebound bursts. Relay cells might choose between excitatory and inhibitory feedback according to the overall noise in the ensemble input.

In this view thalamus behaves like part of a predictive (Kalman) filter, relaying or blocking individual sensory signals based on a specific prediction and its relative confidence. Such a function is broadly consistent with many known features of thalamus and cortex--including recent recordings from free-viewing monkeys--and allows several further experimental tests.

Tony Zador
Tue Oct 22 16:34:57 PDT 1996