[edit]
An Optimal Reduction of TV-Denoising to Adaptive Online Learning
Proceedings of The 24th International Conference on Artificial Intelligence and Statistics, PMLR 130:2899-2907, 2021.
Abstract
We consider the problem of estimating a function from n noisy samples whose discrete Total Variation (TV) is bounded by Cn. We reveal a deep connection to the seemingly disparate problem of \emph{Strongly Adaptive} online learning [Daniely et al 2015] and provide an O(nlogn) time algorithm that attains the near minimax optimal rate of ˜O(n1/3C2/3n) under squared error loss. The resulting algorithm runs online and optimally \emph{adapts} to the \emph{unknown} smoothness parameter Cn. This leads to a new and more versatile alternative to wavelets-based methods for (1) adaptively estimating TV bounded functions; (2) online forecasting of TV bounded trends in time series.