A1036
Title: The proximal distance principle: Algorithms and applications
Authors: Alfonso Landeros - University of California, Riverside (United States) [presenting]
Abstract: Statistical methods often involve solving optimization problems. The addition of constraints, either to enforce a hard requirement in estimation or to regularize solutions, complicates matters. Fortunately, the rich theory of convex optimization provides ample tools for devising novel methods, especially when there is tension between theory and computational demands in a high-dimensional setting. A distance-to-set penalty strategy is discussed as a general approach to solving constrained estimation problems. Special emphasis is given to sparsity constraints for variable selection, which compromise between exhaustive combinatorial searches and shrinkage penalties. Examples drawn from life science applications vividly illustrate the ease of incorporating structure into estimators within the proximal distance framework.
