Professor Ken Miller

Departments of Physiology and Otolaryngology, University of California at San Francisco
Email: ken@phy.ucsf.edu
Web: click here

My lab's interests focus on understanding the cerebral cortex. We use theoretical and computational methods, and theoretically motivated experimental methods, to unravel the circuitry of the cerebral cortex, the rules by which this circuitry develops or "self-organizes", and the computational functions of this circuitry.

One goal of the lab is to understand the role of activity-dependent, "correlation-based" mechanisms of synaptic plasticity in determining cortical structure and function (see lab publications on Models of Neural Development, below). Under these mechanisms, synaptic change appears to follow a rule like that proposed by Hebb in 1949: a synapse is strengthened when pre- and postsynaptic activations are correlated. We have analyzed cortical development in the presence of such plasticity. One prominent feature of visual cortical development is the formation of ocular dominance columns. These are alternating patches of cortical cells that receive input only from the left eye or only from the right eye. The left- and right-eye inputs segregate, beginning from an initially intermixed condition, through an activity-dependent synaptic competition. We have predicted the conditions under which input neural activity will lead to such segregation, and the size of the resulting patches. Another feature of visual cortex is the tuning of the cells to respond to light-dark borders of a particular orientation. Our analysis revealed that the development of such orientation selectivity can be explained by a correlation-based competition between ON-center and OFF-center inputs to the visual cortex, very much like the left-eye/right-eye competition that leads to ocular dominance column formation but in a different parameter regime. More recently we have addressed the combined development of ocular dominance and orientation selectivity, showing how the orientation preferences of the two eyes can become matched despite the tendency of the two eyes to segregate from one another. All of these models make strong, testable predictions as to the pattern of correlations that must exist among the activities of inputs to cortex during development, if the mature cortical structure arises by correlation-based rules. In addition, our hypothesis as to the mechanism of matching of the two eye's orientation preferences leads to testable predictions for the relationship between the two eye's receptive fields in mature visual cortical cells, and explains existing observations of the distribution of best stimulus disparities in these cells.

Another goal of the lab is to develop realistic and testable models of mature cortical circuitry (see lab publications on Models of Neuronal Integration and Circuitry, below). We have developed improved simple models of cortical excitatory cells and shown how these naturally account for the high variability of cortical responses. We have developed a candidate circuit model to explain the full response properties of cortical cells in layer 4 (the input-recipient layer) of cat primary visual cortex, addressing the invariance of orientation tuning under changes in stimulus contrast and a variety of other response properties. The model makes a number of predictions, notably as to the response properties of inhibitory neurons in layer 4. This circuit model involves "correlation-based intracortical circuitry", and thus closely connects with the studies described above of cortical development: we have recently shown how the candidate circuit can itself arise from development under correlation-based plasticity rules. We are continuing to investigate the properties of the layer 4 circuit, and also intend to proceed to other layers.

Finally, we have established experimental methods for the study of the simultaneous activity of many neurons in visual cortex, using the "tetrode" method of recording (see lab publications on Experimental Results, below). Experiments applying these methods in cat visual cortex and LGN (the nucleus providing visual input to cortex) are underway. These will serve both to inform and to test the models. Simultaneous recording from many neurons also provides a basis for other theoretical studies, such as the analysis of the cortical coding and representation of sensory information; our analysis of coding in LGN indicates that neurons have temporal response precision of a few milliseconds and can code more than 3 bits/spike.