Dr Song Liu’s paper ‘Two-sample inference for high-dimensional Markov networks’ has been accepted by the Journal of the Royal Statistical Society (Series B).
Graphical models are probabilistic models that encode dependencies among random variables. By comparing two graphical models under two different settings, we gain important insights about our data: such as how gene networks react to certain stimuli and how the stock market responds to a social event. This paper proposes an inference method that identifies sparse changes between two graphical models using two datasets. Earlier methods work on high dimensional graphical models, but the biases introduced by the sparsity inducing regularizer makes it challenging to specify confidence intervals. In contrast, the proposed estimator is asymptotically normal, which allows us to specify confidence intervals for our detected changes. Therefore, the estimated model is much more interpretable.
A preprint of the paper is available here
