A1184
Title: Recursive nonparametric estimation: Principles, methods and applications
Authors: Man Fung Leung - University of Illinois Urbana-Champaign (United States) [presenting]
Kin Wai Chan - The Chinese University of Hong Kong (Hong Kong)
Abstract: Existing long-run variance estimators face a dilemma between mean squared error, time complexity, and space complexity. A conceptual decomposition will be presented to understand this phenomenon. The new insights allow us to improve existing works, but further efficient estimators in a principle-driven way are characterized. The asymptotic theory and simulations show that this new approach leads to online estimators with a lower mean squared error. It is also discussed practical enhancements such as mini-batch and automatic updates. Encouraging finite-sample results are illustrated in online change point detection, stochastic approximation, and Markov chain Monte Carlo convergence diagnosis.