Using Brain Connectivity and Machine Learning to Understand Skill Learning
Overview
Machine learning approaches offer an important advantage over classical approaches to statistical prediction by preventing overfitting, thus creating models that are more generalizable and likely to replicate. In the current presentation, I will introduce a simple to use machine learning package that we created to analyze neuroimaging data, and I demonstrate how we’ve applied it to real data. Acquisition of motor skills is thought to depend highly on interaction between the motor system and subcortical structures. In our demonstration, we used this machine learning package to examine how functional brain network connectivity in the subcortical and motor networks predict initial performance on a complex motor task, as well as training-induced improvements following 20 hours of training. By contrasting the predictive success of different types of feature sets, we explore how the connectivity both within and between these networks contain differential information for predicting individual performance and learning rate. We also interrogate sets of features that are most reliably included in the models with high predictive accuracy, and discuss what these highly successful feature sets imply for learning and performance of complex motor tasks.
Meeting Info
- Date: Feb 3, 2016
- Time: 12:00 PM
- Location: 2000 NCSA
- Topic: Using Brain Connectivity and Machine Learning to Understand Skill Learning
##Bio:
Aki Nikolaidis investigates how brain networks, metabolism, and structure contribute to individual differences in executive function, intelligence, and learning rates. His work has also focused on understanding how cognitive training drives brain plasticity and improvements in cognition. More broadly, he is interested in using cutting edge statistical methodology to understand how patterns of large scale brain plasticity drive forms of learning that generalize across multiple contexts.