A number of recent publications have implicated the basal ganglia, in particular the caudate nucleus and the substantia nigra, in certain forms of implicit learning. From a computational point of view, the medium spiny cells of the caudate are very well suited (perhaps uniquely suited) for playing a central role in learning to detect complex, difficult to verbalize "rules" about when certain behaviors would yield desirable consequences. Modeling work in our lab invovles constructing anatomically and physiologically plausible network models of the circuits surrounding these cells, with an eye toward perceptual categorization and instrumental conditioning. The work is largely motivated by the perceptual categorization theories and methodologies of F. G. Ashby at UC Santa Barbara, with whom we actively collaborate.
In concert with the modeling work on the basal ganglia, we are also involved in both human and animal investigations of behavioral phenomena surrounding perceptual categorization. In particular, we have begun conducting human perceptual learning studies on auditory and mixed-modality (audio-visiual) stimuli using the randomization technique associated with Ashby's General Recognition Theory. Our observation, following Ashby and colleague's work, is that in the behavioral situations employed in these tasks, human subjects gradually learn to perform optimally (defined in information-theoretic terms) but can not accurately verbalize the rules they use to asign category labels to individual stimuli. We have begun to bring this task to animal subjects (rats) in an instrumental learning paradigm.
Our lab is also involved in investigations of manual reaching to visually specified targets, both of computational modeling and empirical natures. The primary empirical observation is that in visually feedforward conditions (i..e., when subjects must reach to a target, but when vision of the target or reaching hand is precluded immediately as the reach is begun), subjects are only accurate when they can see the results of their reach after the reach is completed. When visually feedforward reaches are not followed by such feedback, reaches are systematically and substantially innaccurate. The spatial nature of these errors is stable for individual subjects over a short period of time (e.g., a single 1-hour experimental session), but slowly drifts over longer periods of time (e.g., the direction of the error can change as much as 180 deg. over the course of several weeks). These results have implications for our understanding of the planning processes that preceed even the simplest voluntary actions.