cold springs harbor laboratory
churchland lab



Study multisensory decision-making in rats and humans.
We present both rat and human subjects with ambiguous stimuli about which they make decisions. The stimuli consist of visual flashes (left, top row), auditory tones (left, bottom row), or both. Subjects must make a judgment about the repetition rate of the stimulus. To determine whether subjects benefit from multisensory integration, we compare the speed and accuracy of the single sensory and the multisensory conditions.

Uncover the neural mechanisms of multisensory decision-making by measuring the responses of neurons in parietal cortex in behaving rodents.
We collect electrophysiological responses from single neurons in parietal cortex. Parietal cortex is known to receive input from both auditory and visual sensory areas and is known to play a key role in decision-making. By comparing the electrophysiological responses of parietal neurons when the animals are engaged in single sensory vs. multisensory decisions, we can explore the nature of the cortical circuitry that makes it possible to support the behavioral improvement in multisensory integration.


Use behavioral and physiological data to evaluate theoretical models.
Behavior and physiology from decision tasks about visual motion have been consistent with a class of models known as Bounded Accumulation (Laming, 1968; Link and Heath, 1975). In these models, random samples of evidence are accumulated over time up to a threshold level or bound. Although this class of models has been successful for visual motion decisions, it is not yet known whether it will generalize to multisensory decisions. Our working hypothesis is that the models will be easily extendable to multisensory decisions: instead of accumulating random samples of evidence only across time, the models can be designed to accumulate random samples of evidence across modalities. We are in the process of testing whether this class of models can accurately predict the speed and accuracy of decisions, and whether its predictions are compatible with electrophysiological observations. Other kinds of models that are currently under consideration are models of Bayesian inference, or attractor models.