Professor David Redish

This research entails studying the information processing that underlies behavior during complex decisions. The group is currently using MSI for two projects:

• Training animals (mice, rats, humans) on complex behavioral tasks. From the rats, they record large ensembles of neural activity. The group's datasets tend to be relatively large (1 GB per session is not unusual and a total dataset from multiple animals and multiple days can be 10 GB to 100 GB). Moreover, analyzing an experimental question can be extremely computationally intensive, requiring MSI supercomputer resources. Typical questions address how information is represented and how it changes as behavior changes. This typically requires a decoding operation in which the researchers calculate the relationship between neural firing and behavioral variables in one situation and then using this calculation along with a new dataset of neural firing to predict information during behavior. For example, they found that when rats come to choices they sometimes deliberate over those choices, representing the alternate options in one neural structure and evaluating the potential options in another. Other times, when rats come to choices, they use a different neural system to represent the current sensory situation, releasing a well-learned motor sequence. Calculating the decoded operation can be computationally intensive.
• Computational models of neural systems. These models entail hundreds or thousands of simulated interacting neurons. These models allow the researchers to test theoretical hypotheses about how neural systems might be computing information. For example, the group has built a model of how pharmacological effects on neural receptors can produce behaviors observed in schizophrenic patients. The researchers can thereby explore predictions of theories translating from low-level neurophysiology to high-level behavior.