These are the codes we have developed for our research projects. Please contact me if you need help with using them.
We have developed a new dimensionality reduction method using Grassmann HMC. This is a joint work with Andrew Holbrook and Alexander Vandenberg.
A Dynamic Bayesian Model for Neuronal Interactions
We have developed a dynamic Bayesian method (published in JASA) for analyzing multiple neurons. This is a joint work with Bo Zhou, Sam Behseta, Hernando Ombao, and David Moorman.
Detecting Synchrony Among Multiple Neurons
We have developed a new approach for detecting multiway synchrony among neurons. For this project, I worked with my students, Bo Zhou and Shiwei Lan, and my collaborators, Sam Behseta, Hernando Ombao, and David Moorman.
Wormhole Hamiltonian Monte Carlo (WHMC)
My student, Shiwei Lan, and I have worked on this project with Professor Jeff Streets. Our method can sample from multimodal distributions even in high dimensions. You can obtain the codes and learn more about this method from Shiwei Lan's website.
Lagrangian Monte Carlo (LMC)
Shiwei Lan, Vassilios Stathopoulos, Mark Girolami and I have have developed a new algorithm called Lagrangian Dynamical Monte Carlo. This is a fully explicit integrator for Riemannian Manifold Monte Carlo (Girolami and Calderhead, 2011). We show that our method is in fact equivalent to transforming Riemannian Hamiltonian dynamics to Lagrangian dynamics. HYou can obtain the codes and learn more about this method from Shiwei Lan's website.
Split Hamiltonian Monte Carlo (Split HMC)
With Radford Neal and Shiwei Lan, we have proposed a new approach to improve computational efficiency of Hamiltonian Monte Carlo (HMC). Our approach is based on "splitting" the Hamiltonian such that much of the movement around the state space is performed at low computational cost.
Nonlinear models using Dirichlet process mixtures
We introduced a new nonlinear method based on modeling the joint distribution of response and covariates using the Dirichlet process mixtures of linear models. This was a joint work with Radford Neal.
These are some files related to our work (with Radford Neal) on classification models when classes have a hierarchical structure.
MNL: This is a simple Bayesian multinomial model (i.e., taking classes as unrelated entities)
treeMNL: This models hierarchical classes using nested multinomial logit models.
corMNL: This is our proposed method that takes the hierarchy as a prior.
makeTree: This code gets a matrix of classes and returns a tree structure.
Bayesian Relevance Determination (BRD)
Wes Johnson and I have developed a new nonparametric Bayesian method to divide genes into several subgroups according to their degree of relevance to the outcome of interest (i.e., disease status).
2224 DBH, UC Irvine, CA 92697
babaks at uci dot edu