Title: Sparse regression and structure hunting in the periodogram analysis of unequally spaced time series
Authors: Ivan Mizera - University of Alberta (Canada) [presenting]
Li Zhang - University of Alberta (Canada)
Abstract: The periodogram methodology is revised along the classical lines, with special attention to (substantially) unequally spaced time series. On the basis of this revision, we propose modifications along the lines of various techniques of sparse regression. Among those figure prominently l1/l2 regularization solved via second-order convex programming; but we also consider possible ramifications, in particular in view of possible theoretical objections to the application of l1/l2 regularization in this specific setting. We pay particular attention to the choice of tuning parameters determining the correct amount of regularization leading to the appropriate model selection: we try several recent structure-oriented approaches (like stability selection, for instance), and propose few others.