Research
I study how the brain learns to perform new movements. To this end, I record kinematic behaviour of study participants as they learn to perform new movements in different ways. My hope is to be able to apply what I learn about motor learning in rehabilitation of patients with impaired motor function.
Here is a short summary of my research projects
Sensory cancellation is one of the most remarkable mechanisms our nervous system uses to distinguish self-generated action from environmentally-generated actions, thus allowing for environmental signals to be processed effectively. A classic example is the inability to tickle onself because the motor system predicts the sensation beforehand and cancels the overall sensation. In this project, we showed that such sensory cancellation is highly accurate, in that the motor system can predict very small movements with high fidelity. This means that even though our actions are not always precise, our prediction about the consequences of those actions are very precise
When learning to perform a new action (e.g., making a tennis serve), one has information about final motor performance as well as continuous error signal in the form of the ball’s trajectory. Both these error signals are useful predictors of motor adaptation. In this work, we asked which of these two error signals drives the short-term component of learning. Our results show that it is the motor performance error that drives motor adaptation in the short-term
Mental health disorders affect a large population worldwide, yet the resources for online help remain scarce. In a collaboration with clinicians at the Beth Israel Deaconess medical center, I am exploring methods to predict relapse episodes in patients with Schizophrenia through data collected via a smartphone app called mindLAMP. I use background data - such as GPS or call/text log to predict survey responses about mental health status
Perceptual decision-making tasks involve transforming continuous sensory inputs into discrete decisions and continuous actions (Parr & Friston, 2018). This transformation occurs across a large network of brain areas. In mice performing such a task, neurons encoding choice are localized in forebrain and midbrain regions, while neurons encoding action initiation are distributed across the brain (Steinmetz et al. 2019). Whether the representation of brain-wide neural activity during such a transformation evolves in a continuous or discrete manner remains unclear (VanRullen & Koch, 2003). Hidden Markov models (HMM) can be used to assess neural dynamics when latent neural states are assumed to be discrete. Here, we compare this approach with that describing smooth evolution of states by decoding behavior following either HMM, or independent component analysis (ICA) using data collected by Steinmetz et al (2019).
