Statistics and Biostatistics seminar series
Babak Shahbaba
University of California, Irvine (UCI)
Room: M3 3127
Of rodents and humans: a framework to capture latent representations and dynamics in brain activity data across species
In this talk, I will discuss some new statistical machine learning methods to help identify the sequences of brain activity states that result in specific decisions. Our proposed framework is designed to facilitate visualization, exploration, modeling, and interpretation of the underlying (latent) patterns in high dimensional, multiscale, and multimodal neural data. To achieve this goal, we have developed a novel deep learning representation and integration method to extract latent states and trajectories from the brain activity. Additionally, we have developed a graph neural network to capture the underlying structure of brain dynamics. We validate our methods using a one-step decision task (one response for each stimulus), which offers a simpler platform to quantify sequences of brain activity states and structures. I will then discuss their extensions to multi-step planning tasks as well as to cross-species multimodal data. The long-term goal of this research is to advance our fundamental understanding of the neural mechanisms supporting goal-directed decisions, and to guide follow-up experiments investigating how these principles change in cognitive disorders.