Machine learning approaches towards understanding movement planning in naturalistic settings

Public Deposited

A central question in neuroscience is how the brain plans movements. Here, I apply neural data analysis and machine learning methods to better understand both eye and arm movement planning, in particular focusing on naturalistic settings. First, I built encoding models to investigate the factors that led to neural activity in macaque Frontal Eye Field (FEF) during a natural scene search task (Ch. 2,3,4). One central finding was that FEF neurons did not represent task-relevant visual features within natural scenes. Another central finding was that separate populations of neurons represented preliminary and definitive plans for movement. The neurons that represented preliminary plans represented the probabilities of potential upcoming saccades. I found similar characteristics in dorsal premotor cortex, where populations of neurons represented the probability distributions of possible upcoming reaches (Ch. 5). Finally, I compared many different methods for neural decoding to demonstrate that modern machine learning methods lead to performance improvements, even for limited amounts of data (Ch. 6). Overall, I have provided insights into neural activity across a wide range of motor behaviors in more naturalistic settings, and have demonstrated the value of using machine learning methods within neuroscience.

Last modified
  • 02/22/2019
Date created
Resource type
Rights statement