Abstract:
|
In this talk, we consider several signal detection problems that share a null vs. spiked alternative setup. This general setup includes, but is not limited to, Erdos—Renyi random graphs vs. planted partition models, white noise models vs. finite Gaussian mixtures, and white noise vs. spiked covariance models. We discuss the asymptotic normality of log-likelihood ratios against local alternatives and polynomial time detection algorithms based on linear spectral statistics.
|