Tue, Oct. 10 4:10 pm, FLO 100 Refreshments at 4:00 pm

In this talk, I will provide an overview of both frequentist and Bayesian approaches to the well-known problem of estimating a sparse n-dimensional unknown mean vector θ = (θ1, …, θn), with entries corrupted by Gaussian white noise. This simple framework is the basis for many real-world problems such as image reconstruction and gene identification. Sparsity means that nearly all of the entries in the true underlying θ are zero, and one of the challenges for scientists and engineers is identifying and estimating the true signals (i.e. the actual non-zero parameters) in the noisy data.

In this talk, I will also talk about a new type of prior we have invented known as the inverse gamma-gamma (IGG) prior and how it can be used to identify and estimate signals. I will discuss the IGG’s theoretical properties and demonstrate its finite sample performance.