Research Areas
Learning
Learning to do useful, goal oriented behavior, whether it is playing a violin or tying our shoes takes time, effort and practice. We study corresponding changes in the brain as a function of multiple time scales, ranging from minutes to years. This is reflected in the measured depth, resilience and specificity of the resulting skills. The research is extended to understanding brain reorganization and recovery in the setting of neural injury and degeneration in patients. The goal is to develop new methods of accelerating and amplifying systems that support the capacity to acquire compensatory behaviors that still achieve useful motor outcomes and lead to functional independence. Our learning paradigms in humans are linked to single unit recordings, optical imaging and pharmacological manipulation in non-human primates doing similar behaviors and computational modeling with collaborators at University of Pittsburgh, Northwestern University and UCSB.
Control
Contemporary theories frame motor control in terms of optimal control theory. A central aspect of this theory is a state-space framework that integrates desired motor outcomes in terms of action goals with ongoing motor behavior, the predicted sensory consequences of this action and feedback from both peripheral sources and motor commands. We focus on the role of parietal cortex for generating a goal-based representation that supports state estimation. Goal, feedback and output perturbations are manipulated to characterize neural substrates of state estimation by fMRI, high-density EEG and TMS.
Action Understanding
We have characterized a mosaic of cortical areas that are consistently involved in the recognition of actions performed by other people. The contribution of areas in the action observation is not uniform, and varies as a function of what aspect of an action is being decoded. There is evidence for a nested, relative functional hierarchy that distinguishes lower level kinematic features from object-centered actions and their resultant outcomes. The work provides evidence that much (but not all) of what is understood about others behavior is based on embodied cognition acquired through our own physical experience. Ongoing work aims to understand how prosody of movement, object features and cross agent actions are represented. Repetition suppression (fMRI adaptation) is the primary approach to make these functional distinctions. Cross-agent studies with repetition suppression are used to characterize the degree to which there are shared neural substrates for action generation and action understanding.
Action Organization
Making a cup of coffee in an unfamiliar hotel room requires an ability to map existing knowledge structures about objects into a novel action schedule. This requires an ability to explicitly plan and flexibly adapt action elements into an efficient solution. While this process can be described with formal hierarchical models of anticipatory planning, it is is also striking that a few days later the same action has already become an automated action requiring minimal thought. We are investigating the neural substrates that support this transition from formal hierarchical problem solving to programmatic control of naturalistic behavior with repetition suppression and transcranial magnetic stimulation. The work is linked to patients with apraxia and disorders of naturalistic action in a multi-center collaborative study with investigators at the Moss Rehabilitation Center, Einstein University, University of Oregon and Princeton.
Individual Differences
We participate in a larger scale effort at UCSB to characterize how individual differences in brain structure and function are reflected in individual differences in performance. This is done with studies of spatial attention, visuomotor control and decision-making.
New Methods
- We collaborate with the Department of Physics to develop a computational framework for integrating fMRI and EEG data from the same subject. This is being extended to develop methods to analyze EEG data acquired simultaneous with fMRI.
- Pattern classification schemes based on non-negative matrix factorization are being developed to classify structural MRI scans from different subjects into specific populations.
- Computational methods from optimal control theory are used to characterize motor performance, with the goal of improving model prediction when physiologic data is available.
- Sensitivity of learning as a function of time of day, chronotype and sleep pressure and being integrated into a unified computational framework. This will allow for a better characterization of individual differences in skill learning based on chronobiology.