Ph.D. Research Proficiency Exam: Yididiya Nadew

Yididiya Nadew
Thursday, May 16, 2024 - 9:00am
223 Atanasoff Hall
Event Type: 

Latent Variable Models For Neuronal Spike Trains

Modern recording techniques have enabled neuroscientists to simultaneously record the spiking activity of 10s-100s of neurons. Such “spike train” data is noisy and high-dimensional. To ease visualization and interpretation, researchers have adopted latent variable models (LVMs), in which a few variables are learned that summarize the data well, for lowerdimensional representations. One popular class of these LVMs, Gaussian processes factor analysis (GPFA) models, recover smooth latent trajectories. However, due to the nonconjugate form of the likelihood, Bayesian inference of this model remained intractable. Prior works rely on either black-box inference techniques, numerical integration or polynomial approximations of the likelihood to handle intractability. To overcome this challenge, we build a conditionally-conjugate Gaussian process factor analysis (ccGPFA) model resulting in both analytically and computationally tractable inference. In particular, we develop a novel data augmentation based method that renders the model conditionally conjugate. Consequently, our model enjoys the advantage of simple closed-form updates using a variational EM algorithm. We conducted experiments on mouse and primate data to demonstrate the effectiveness of our model compared to existing ones.

Committee: Chris Quinn (major professor), Hongyang Gao, Wallapak Tavanapong, Jin Tian and Mengdi Huai